The formula may be simple. The meaning belongs to the domain.
A reusable feature is more than a number: it declares what was observed, when it became available, how it was calculated and what it must not be interpreted to mean. Compare twelve domains using the same learning sequence.
vital signs · laboratory · encounter Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
patient_id, event_time, HR=96, HR_prev=82, SpO2=93, temp=38.1Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
SDNN, RMSSD จากระยะ R–R
Attach timestamps, parameters and code version.
WORKED EXAMPLE
HR_delta = 96−82 = 14 bpm; fever = 1; hours_since_medication = 6A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
heart-rate_variability__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
O(n) ต่อผู้ป่วย; rolling 30 วันบน 1M events ≈ วินาที–นาทีใน columnar engineActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ต้องยึด event time, หน่วยวัด และช่วงที่ข้อมูลพร้อมใช้จริง ห้ามนำผลตรวจที่ออกภายหลังมาใช้ย้อนหลัง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
vital signs · laboratory · encounter Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
patient_id, event_time, HR=96, HR_prev=82, SpO2=93, temp=38.1Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
แนวโน้มชีพจร ความดัน อุณหภูมิ SpO₂
Attach timestamps, parameters and code version.
WORKED EXAMPLE
HR_delta = 96−82 = 14 bpm; fever = 1; hours_since_medication = 6A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
news-like_trend__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
O(n) ต่อผู้ป่วย; rolling 30 วันบน 1M events ≈ วินาที–นาทีใน columnar engineActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ต้องยึด event time, หน่วยวัด และช่วงที่ข้อมูลพร้อมใช้จริง ห้ามนำผลตรวจที่ออกภายหลังมาใช้ย้อนหลัง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
vital signs · laboratory · encounter Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
patient_id, event_time, HR=96, HR_prev=82, SpO2=93, temp=38.1Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ค่าปัจจุบัน − ค่าก่อนหน้า
Attach timestamps, parameters and code version.
WORKED EXAMPLE
HR_delta = 96−82 = 14 bpm; fever = 1; hours_since_medication = 6A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
lab_delta__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
O(n) ต่อผู้ป่วย; rolling 30 วันบน 1M events ≈ วินาที–นาทีใน columnar engineActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ต้องยึด event time, หน่วยวัด และช่วงที่ข้อมูลพร้อมใช้จริง ห้ามนำผลตรวจที่ออกภายหลังมาใช้ย้อนหลัง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
vital signs · laboratory · encounter Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
patient_id, event_time, HR=96, HR_prev=82, SpO2=93, temp=38.1Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
จำนวนโรคร่วมในช่วงเวลาที่กำหนด
Attach timestamps, parameters and code version.
WORKED EXAMPLE
HR_delta = 96−82 = 14 bpm; fever = 1; hours_since_medication = 6A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
comorbidity_count__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
O(n) ต่อผู้ป่วย; rolling 30 วันบน 1M events ≈ วินาที–นาทีใน columnar engineActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ต้องยึด event time, หน่วยวัด และช่วงที่ข้อมูลพร้อมใช้จริง ห้ามนำผลตรวจที่ออกภายหลังมาใช้ย้อนหลัง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
vital signs · laboratory · encounter Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
patient_id, event_time, HR=96, HR_prev=82, SpO2=93, temp=38.1Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
dose × duration และเวลาหลังได้รับยา
Attach timestamps, parameters and code version.
WORKED EXAMPLE
HR_delta = 96−82 = 14 bpm; fever = 1; hours_since_medication = 6A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
medication_exposure__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
O(n) ต่อผู้ป่วย; rolling 30 วันบน 1M events ≈ วินาที–นาทีใน columnar engineActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ต้องยึด event time, หน่วยวัด และช่วงที่ข้อมูลพร้อมใช้จริง ห้ามนำผลตรวจที่ออกภายหลังมาใช้ย้อนหลัง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DNA · RNA · protein · omics Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
sequence=ACGCGTATGC; expression=120; cohort_mean=80; sd=20Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
(G+C)/sequence length
Attach timestamps, parameters and code version.
WORKED EXAMPLE
GC=6/10=0.60; expression_z=(120−80)/20=2.0A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
gc_content__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
composition O(n); k-mer O(n); alignment โดยทั่วไป O(nm); RNA folding แบบคลาสสิก O(n³)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ระบุ genome build, strand, transcript version, normalization และ tool version ทุกครั้ง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
k-mer frequency
จำนวนคำยาว k เช่น 3-mer/6-merแทนรูปแบบเฉพาะที่โดยไม่ต้อง alignment
＋
WHAT IT MEASURES AND WHY
แทนรูปแบบเฉพาะที่โดยไม่ต้อง alignment
SOURCE INPUT
DNA · RNA · protein · omics Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
sequence=ACGCGTATGC; expression=120; cohort_mean=80; sd=20Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
จำนวนคำยาว k เช่น 3-mer/6-mer
Attach timestamps, parameters and code version.
WORKED EXAMPLE
GC=6/10=0.60; expression_z=(120−80)/20=2.0A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
k-mer_frequency__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
composition O(n); k-mer O(n); alignment โดยทั่วไป O(nm); RNA folding แบบคลาสสิก O(n³)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ระบุ genome build, strand, transcript version, normalization และ tool version ทุกครั้ง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DNA · RNA · protein · omics Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
sequence=ACGCGTATGC; expression=120; cohort_mean=80; sd=20Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ความยาวช่วงเปิดการอ่าน
Attach timestamps, parameters and code version.
WORKED EXAMPLE
GC=6/10=0.60; expression_z=(120−80)/20=2.0A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
orf_length__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
composition O(n); k-mer O(n); alignment โดยทั่วไป O(nm); RNA folding แบบคลาสสิก O(n³)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ระบุ genome build, strand, transcript version, normalization และ tool version ทุกครั้ง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DNA · RNA · protein · omics Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
sequence=ACGCGTATGC; expression=120; cohort_mean=80; sd=20Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
MFE, paired fraction, stem/loop
Attach timestamps, parameters and code version.
WORKED EXAMPLE
GC=6/10=0.60; expression_z=(120−80)/20=2.0A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
secondary_structure__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
composition O(n); k-mer O(n); alignment โดยทั่วไป O(nm); RNA folding แบบคลาสสิก O(n³)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ระบุ genome build, strand, transcript version, normalization และ tool version ทุกครั้ง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DNA · RNA · protein · omics Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
sequence=ACGCGTATGC; expression=120; cohort_mean=80; sd=20Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
expression เทียบกลุ่มอ้างอิง
Attach timestamps, parameters and code version.
WORKED EXAMPLE
GC=6/10=0.60; expression_z=(120−80)/20=2.0A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
expression_z-score__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
composition O(n); k-mer O(n); alignment โดยทั่วไป O(nm); RNA folding แบบคลาสสิก O(n³)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ระบุ genome build, strand, transcript version, normalization และ tool version ทุกครั้ง One feature is one piece of evidence—not the complete state of a person, machine, market or system.
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
Log return
ln(Pt/Pt−1)ผลตอบแทนที่บวกสะสมตามเวลาได้
＋
WHAT IT MEASURES AND WHY
ผลตอบแทนที่บวกสะสมตามเวลาได้
SOURCE INPUT
price · volume · quote · fundamentals Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
close_prev=100, close=103, bid=102.8, ask=103.2, debt=60M, equity=40MUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ln(Pt/Pt−1)
Attach timestamps, parameters and code version.
WORKED EXAMPLE
return=ln(103/100)=2.956%; spread=0.4/103=0.388%; D/E=1.5A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
log_return__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling feature O(T); 5,000 assets × 10 years daily ≈ วินาที–นาทีแบบ vectorizedActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ปรับ split/dividend, calendar และ survivorship bias; Indicator ไม่ใช่คำสั่งซื้อขาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
price · volume · quote · fundamentals Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
close_prev=100, close=103, bid=102.8, ask=103.2, debt=60M, equity=40MUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
SD ของ return ใน window
Attach timestamps, parameters and code version.
WORKED EXAMPLE
return=ln(103/100)=2.956%; spread=0.4/103=0.388%; D/E=1.5A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
realized_volatility__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling feature O(T); 5,000 assets × 10 years daily ≈ วินาที–นาทีแบบ vectorizedActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ปรับ split/dividend, calendar และ survivorship bias; Indicator ไม่ใช่คำสั่งซื้อขาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
03
RSI
อัตราส่วน average gain/lossบอกตำแหน่ง momentum ในกรอบที่เลือก
＋
WHAT IT MEASURES AND WHY
บอกตำแหน่ง momentum ในกรอบที่เลือก
SOURCE INPUT
price · volume · quote · fundamentals Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
close_prev=100, close=103, bid=102.8, ask=103.2, debt=60M, equity=40MUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
อัตราส่วน average gain/loss
Attach timestamps, parameters and code version.
WORKED EXAMPLE
return=ln(103/100)=2.956%; spread=0.4/103=0.388%; D/E=1.5A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rsi__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling feature O(T); 5,000 assets × 10 years daily ≈ วินาที–นาทีแบบ vectorizedActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ปรับ split/dividend, calendar และ survivorship bias; Indicator ไม่ใช่คำสั่งซื้อขาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Relative spread
(ask−bid)/midประมาณ quoted liquidity ณ เวลาหนึ่ง
＋
WHAT IT MEASURES AND WHY
ประมาณ quoted liquidity ณ เวลาหนึ่ง
SOURCE INPUT
price · volume · quote · fundamentals Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
close_prev=100, close=103, bid=102.8, ask=103.2, debt=60M, equity=40MUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
(ask−bid)/mid
Attach timestamps, parameters and code version.
WORKED EXAMPLE
return=ln(103/100)=2.956%; spread=0.4/103=0.388%; D/E=1.5A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
relative_spread__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling feature O(T); 5,000 assets × 10 years daily ≈ วินาที–นาทีแบบ vectorizedActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ปรับ split/dividend, calendar และ survivorship bias; Indicator ไม่ใช่คำสั่งซื้อขาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Debt-to-equity
total debt/equityอ่านโครงสร้างเงินทุนของกิจการ
＋
WHAT IT MEASURES AND WHY
อ่านโครงสร้างเงินทุนของกิจการ
SOURCE INPUT
price · volume · quote · fundamentals Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
close_prev=100, close=103, bid=102.8, ask=103.2, debt=60M, equity=40MUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
total debt/equity
Attach timestamps, parameters and code version.
WORKED EXAMPLE
return=ln(103/100)=2.956%; spread=0.4/103=0.388%; D/E=1.5A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
debt-to-equity__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling feature O(T); 5,000 assets × 10 years daily ≈ วินาที–นาทีแบบ vectorizedActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ปรับ split/dividend, calendar และ survivorship bias; Indicator ไม่ใช่คำสั่งซื้อขาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
RMS vibration
sqrt(mean(x²))พลังงานสั่นสะเทือนของเครื่อง
＋
WHAT IT MEASURES AND WHY
พลังงานสั่นสะเทือนของเครื่อง
SOURCE INPUT
sensor · machine state · quality Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 unitsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
sqrt(mean(x²))
Attach timestamps, parameters and code version.
WORKED EXAMPLE
RMS=sqrt(10/4)=1.58; slope=12/20=0.6°C/min; defect_rate=0.6%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rms_vibration__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
window W ต่อ sensor: O(N); FFT feature O(N log N); streaming latency มัก ms–sActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
แยก machine mode และ maintenance period มิฉะนั้น model จะสับสนระหว่างงานหนักกับความเสียหาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Kurtosis
ความหนาหางของสัญญาณช่วยเห็น impulsive fault
＋
WHAT IT MEASURES AND WHY
ช่วยเห็น impulsive fault
SOURCE INPUT
sensor · machine state · quality Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 unitsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ความหนาหางของสัญญาณ
Attach timestamps, parameters and code version.
WORKED EXAMPLE
RMS=sqrt(10/4)=1.58; slope=12/20=0.6°C/min; defect_rate=0.6%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
kurtosis__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
window W ต่อ sensor: O(N); FFT feature O(N log N); streaming latency มัก ms–sActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
แยก machine mode และ maintenance period มิฉะนั้น model จะสับสนระหว่างงานหนักกับความเสียหาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
03
Temperature slope
Δtemperature/Δtimeจับภาวะร้อนขึ้นต่อเนื่อง
＋
WHAT IT MEASURES AND WHY
จับภาวะร้อนขึ้นต่อเนื่อง
SOURCE INPUT
sensor · machine state · quality Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 unitsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
Δtemperature/Δtime
Attach timestamps, parameters and code version.
WORKED EXAMPLE
RMS=sqrt(10/4)=1.58; slope=12/20=0.6°C/min; defect_rate=0.6%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
temperature_slope__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
window W ต่อ sensor: O(N); FFT feature O(N log N); streaming latency มัก ms–sActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
แยก machine mode และ maintenance period มิฉะนั้น model จะสับสนระหว่างงานหนักกับความเสียหาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Cycle-time deviation
cycle−baselineเห็นคอขวดหรือเสื่อมสมรรถนะ
＋
WHAT IT MEASURES AND WHY
เห็นคอขวดหรือเสื่อมสมรรถนะ
SOURCE INPUT
sensor · machine state · quality Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 unitsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
cycle−baseline
Attach timestamps, parameters and code version.
WORKED EXAMPLE
RMS=sqrt(10/4)=1.58; slope=12/20=0.6°C/min; defect_rate=0.6%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
cycle-time_deviation__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
window W ต่อ sensor: O(N); FFT feature O(N log N); streaming latency มัก ms–sActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
แยก machine mode และ maintenance period มิฉะนั้น model จะสับสนระหว่างงานหนักกับความเสียหาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Defect rate
defects/unitsคุณภาพผลผลิตตาม line/shift
＋
WHAT IT MEASURES AND WHY
คุณภาพผลผลิตตาม line/shift
SOURCE INPUT
sensor · machine state · quality Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 unitsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
defects/units
Attach timestamps, parameters and code version.
WORKED EXAMPLE
RMS=sqrt(10/4)=1.58; slope=12/20=0.6°C/min; defect_rate=0.6%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
defect_rate__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
window W ต่อ sensor: O(N); FFT feature O(N log N); streaming latency มัก ms–sActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
แยก machine mode และ maintenance period มิฉะนั้น model จะสับสนระหว่างงานหนักกับความเสียหาย One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DATA EXAMPLE
A small observation
vibration=[1,−2,2,−1]; temp 70→82°C in 20 min; 3 defects/500 units
transaction · customer · product Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
last_purchase=10 days; 4 orders/90d; spend=3200; baseline=100; promo_sales=135Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
recency, frequency, monetary
Attach timestamps, parameters and code version.
WORKED EXAMPLE
R=10, F=4, M=3200; apparent_lift=(135−100)/100=35%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rfm__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
group-by O(N); exact basket pairs O(B·k²); 100M lines ≈ นาทีใน distributed warehouseActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
Promotion lift ที่ไม่มี control ไม่ใช่ causal effect; คืนสินค้าและ identity resolution ต้องถูกรวม One feature is one piece of evidence—not the complete state of a person, machine, market or system.
transaction · customer · product Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
last_purchase=10 days; 4 orders/90d; spend=3200; baseline=100; promo_sales=135Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
จำนวน/มูลค่าต่อใบเสร็จ
Attach timestamps, parameters and code version.
WORKED EXAMPLE
R=10, F=4, M=3200; apparent_lift=(135−100)/100=35%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
basket_size__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
group-by O(N); exact basket pairs O(B·k²); 100M lines ≈ นาทีใน distributed warehouseActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
Promotion lift ที่ไม่มี control ไม่ใช่ causal effect; คืนสินค้าและ identity resolution ต้องถูกรวม One feature is one piece of evidence—not the complete state of a person, machine, market or system.
transaction · customer · product Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
last_purchase=10 days; 4 orders/90d; spend=3200; baseline=100; promo_sales=135Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
สัดส่วนยอดซื้อรายหมวด
Attach timestamps, parameters and code version.
WORKED EXAMPLE
R=10, F=4, M=3200; apparent_lift=(135−100)/100=35%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
category_affinity__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
group-by O(N); exact basket pairs O(B·k²); 100M lines ≈ นาทีใน distributed warehouseActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
Promotion lift ที่ไม่มี control ไม่ใช่ causal effect; คืนสินค้าและ identity resolution ต้องถูกรวม One feature is one piece of evidence—not the complete state of a person, machine, market or system.
transaction · customer · product Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
last_purchase=10 days; 4 orders/90d; spend=3200; baseline=100; promo_sales=135Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ยอดช่วงโปร − baseline
Attach timestamps, parameters and code version.
WORKED EXAMPLE
R=10, F=4, M=3200; apparent_lift=(135−100)/100=35%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
promotion_lift__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
group-by O(N); exact basket pairs O(B·k²); 100M lines ≈ นาทีใน distributed warehouseActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
Promotion lift ที่ไม่มี control ไม่ใช่ causal effect; คืนสินค้าและ identity resolution ต้องถูกรวม One feature is one piece of evidence—not the complete state of a person, machine, market or system.
transaction · customer · product Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
last_purchase=10 days; 4 orders/90d; spend=3200; baseline=100; promo_sales=135Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
เวลาที่สินค้าขาด/เวลาขาย
Attach timestamps, parameters and code version.
WORKED EXAMPLE
R=10, F=4, M=3200; apparent_lift=(135−100)/100=35%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
stockout_exposure__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
group-by O(N); exact basket pairs O(B·k²); 100M lines ≈ นาทีใน distributed warehouseActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
Promotion lift ที่ไม่มี control ไม่ใช่ causal effect; คืนสินค้าและ identity resolution ต้องถูกรวม One feature is one piece of evidence—not the complete state of a person, machine, market or system.
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
Flow duration
last packet−first packetแยก session สั้น/ยาว
＋
WHAT IT MEASURES AND WHY
แยก session สั้น/ยาว
SOURCE INPUT
flow · log · identity · endpoint Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
flow: 20 packets, 12,000 bytes, 4 sec; login failures=12/5minUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
last packet−first packet
Attach timestamps, parameters and code version.
WORKED EXAMPLE
600 bytes/packet; 5 packets/s; failure_velocity=2.4/minA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
flow_duration__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
flow aggregation O(P); entropy O(k); streaming ต่อ event ≈ sub-ms–ms หาก state อยู่ใน memoryActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
รักษา source clock, NAT และ asset role; rarity ไม่เท่ากับ malicious และ alert ต้องมีบริบท One feature is one piece of evidence—not the complete state of a person, machine, market or system.
flow · log · identity · endpoint Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
flow: 20 packets, 12,000 bytes, 4 sec; login failures=12/5minUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
bytes/packet count
Attach timestamps, parameters and code version.
WORKED EXAMPLE
600 bytes/packet; 5 packets/s; failure_velocity=2.4/minA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
bytes_packet__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
flow aggregation O(P); entropy O(k); streaming ต่อ event ≈ sub-ms–ms หาก state อยู่ใน memoryActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
รักษา source clock, NAT และ asset role; rarity ไม่เท่ากับ malicious และ alert ต้องมีบริบท One feature is one piece of evidence—not the complete state of a person, machine, market or system.
flow · log · identity · endpoint Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
flow: 20 packets, 12,000 bytes, 4 sec; login failures=12/5minUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
failure ต่อ user/IP/window
Attach timestamps, parameters and code version.
WORKED EXAMPLE
600 bytes/packet; 5 packets/s; failure_velocity=2.4/minA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
failed-login_velocity__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
flow aggregation O(P); entropy O(k); streaming ต่อ event ≈ sub-ms–ms หาก state อยู่ใน memoryActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
รักษา source clock, NAT และ asset role; rarity ไม่เท่ากับ malicious และ alert ต้องมีบริบท One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Destination entropy
entropy ของปลายทางบอกความกระจายของ scan/beacon
＋
WHAT IT MEASURES AND WHY
บอกความกระจายของ scan/beacon
SOURCE INPUT
flow · log · identity · endpoint Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
flow: 20 packets, 12,000 bytes, 4 sec; login failures=12/5minUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
entropy ของปลายทาง
Attach timestamps, parameters and code version.
WORKED EXAMPLE
600 bytes/packet; 5 packets/s; failure_velocity=2.4/minA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
destination_entropy__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
flow aggregation O(P); entropy O(k); streaming ต่อ event ≈ sub-ms–ms หาก state อยู่ใน memoryActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
รักษา source clock, NAT และ asset role; rarity ไม่เท่ากับ malicious และ alert ต้องมีบริบท One feature is one piece of evidence—not the complete state of a person, machine, market or system.
flow · log · identity · endpoint Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
flow: 20 packets, 12,000 bytes, 4 sec; login failures=12/5minUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
−log frequency(process)
Attach timestamps, parameters and code version.
WORKED EXAMPLE
600 bytes/packet; 5 packets/s; failure_velocity=2.4/minA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rare_process_score__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
flow aggregation O(P); entropy O(k); streaming ต่อ event ≈ sub-ms–ms หาก state อยู่ใน memoryActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
รักษา source clock, NAT และ asset role; rarity ไม่เท่ากับ malicious และ alert ต้องมีบริบท One feature is one piece of evidence—not the complete state of a person, machine, market or system.
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
TF–IDF
term frequency × inverse document frequencyให้น้ำหนักคำที่จำเพาะต่อเอกสาร
＋
WHAT IT MEASURES AND WHY
ให้น้ำหนักคำที่จำเพาะต่อเอกสาร
SOURCE INPUT
document · token · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
term frequency × inverse document frequency
Attach timestamps, parameters and code version.
WORKED EXAMPLE
IDF=ln(100/4)=3.219; TF–IDF=3×3.219=9.657A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
tf_idf__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
TF–IDF O(tokens); transformer โดยคร่าว O(L²d) ต่อ layer; batch GPU ≈ ms–s/documentActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ภาษา การตัดคำ tokenizer และ model version เปลี่ยน feature; sentiment ข้าม domain อาจตีความผิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
document · token · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
token, sentence, syllable statistics
Attach timestamps, parameters and code version.
WORKED EXAMPLE
IDF=ln(100/4)=3.219; TF–IDF=3×3.219=9.657A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
length_readability__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
TF–IDF O(tokens); transformer โดยคร่าว O(L²d) ต่อ layer; batch GPU ≈ ms–s/documentActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ภาษา การตัดคำ tokenizer และ model version เปลี่ยน feature; sentiment ข้าม domain อาจตีความผิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
03
Sentiment score
polarity/confidenceสรุปท่าทีตาม model/domain
＋
WHAT IT MEASURES AND WHY
สรุปท่าทีตาม model/domain
SOURCE INPUT
document · token · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
polarity/confidence
Attach timestamps, parameters and code version.
WORKED EXAMPLE
IDF=ln(100/4)=3.219; TF–IDF=3×3.219=9.657A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
sentiment_score__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
TF–IDF O(tokens); transformer โดยคร่าว O(L²d) ต่อ layer; batch GPU ≈ ms–s/documentActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ภาษา การตัดคำ tokenizer และ model version เปลี่ยน feature; sentiment ข้าม domain อาจตีความผิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
document · token · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
จำนวนและชนิดบุคคล องค์กร สถานที่
Attach timestamps, parameters and code version.
WORKED EXAMPLE
IDF=ln(100/4)=3.219; TF–IDF=3×3.219=9.657A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
named_entities__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
TF–IDF O(tokens); transformer โดยคร่าว O(L²d) ต่อ layer; batch GPU ≈ ms–s/documentActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ภาษา การตัดคำ tokenizer และ model version เปลี่ยน feature; sentiment ข้าม domain อาจตีความผิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Embedding
dense semantic vectorเทียบความคล้ายเชิงบริบท
＋
WHAT IT MEASURES AND WHY
เทียบความคล้ายเชิงบริบท
SOURCE INPUT
document · token · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3Use only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
dense semantic vector
Attach timestamps, parameters and code version.
WORKED EXAMPLE
IDF=ln(100/4)=3.219; TF–IDF=3×3.219=9.657A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
embedding__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
TF–IDF O(tokens); transformer โดยคร่าว O(L²d) ต่อ layer; batch GPU ≈ ms–s/documentActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
ภาษา การตัดคำ tokenizer และ model version เปลี่ยน feature; sentiment ข้าม domain อาจตีความผิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DATA EXAMPLE
A small observation
docs=100; “pipeline” appears in 4 docs; term frequency in doc=3
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
Color histogram
การกระจายค่าสีต่อ channelสรุปสีโดยไม่รักษาตำแหน่ง
＋
WHAT IT MEASURES AND WHY
สรุปสีโดยไม่รักษาตำแหน่ง
SOURCE INPUT
image · region · pixel · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
object area=400 px²; perimeter=100 px; edges=2,000/40,000 pixelsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
การกระจายค่าสีต่อ channel
Attach timestamps, parameters and code version.
WORKED EXAMPLE
edge_density=5%; compactness=4πA/P²=0.503A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
color_histogram__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
histogram O(HW); convolution ≈ O(HWk²CinCout); GPU inference ≈ ms–100ms/imageActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resize, crop, lighting และ camera domain เป็นส่วนหนึ่งของนิยาม Feature ไม่ใช่ preprocessing ที่มองข้ามได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Edge density
edge pixels/all pixelsแทนความซับซ้อนของเส้นขอบ
＋
WHAT IT MEASURES AND WHY
แทนความซับซ้อนของเส้นขอบ
SOURCE INPUT
image · region · pixel · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
object area=400 px²; perimeter=100 px; edges=2,000/40,000 pixelsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
edge pixels/all pixels
Attach timestamps, parameters and code version.
WORKED EXAMPLE
edge_density=5%; compactness=4πA/P²=0.503A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
edge_density__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
histogram O(HW); convolution ≈ O(HWk²CinCout); GPU inference ≈ ms–100ms/imageActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resize, crop, lighting และ camera domain เป็นส่วนหนึ่งของนิยาม Feature ไม่ใช่ preprocessing ที่มองข้ามได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
image · region · pixel · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
object area=400 px²; perimeter=100 px; edges=2,000/40,000 pixelsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
LBP/GLCM contrast, homogeneity
Attach timestamps, parameters and code version.
WORKED EXAMPLE
edge_density=5%; compactness=4πA/P²=0.503A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
texture__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
histogram O(HW); convolution ≈ O(HWk²CinCout); GPU inference ≈ ms–100ms/imageActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resize, crop, lighting และ camera domain เป็นส่วนหนึ่งของนิยาม Feature ไม่ใช่ preprocessing ที่มองข้ามได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Shape descriptors
area, perimeter, compactnessตีความรูปทรงของ region
＋
WHAT IT MEASURES AND WHY
ตีความรูปทรงของ region
SOURCE INPUT
image · region · pixel · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
object area=400 px²; perimeter=100 px; edges=2,000/40,000 pixelsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
area, perimeter, compactness
Attach timestamps, parameters and code version.
WORKED EXAMPLE
edge_density=5%; compactness=4πA/P²=0.503A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
shape_descriptors__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
histogram O(HW); convolution ≈ O(HWk²CinCout); GPU inference ≈ ms–100ms/imageActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resize, crop, lighting และ camera domain เป็นส่วนหนึ่งของนิยาม Feature ไม่ใช่ preprocessing ที่มองข้ามได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
CNN embedding
vector จากชั้นลึกแทนลักษณะภาพระดับสูง
＋
WHAT IT MEASURES AND WHY
แทนลักษณะภาพระดับสูง
SOURCE INPUT
image · region · pixel · embedding Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
object area=400 px²; perimeter=100 px; edges=2,000/40,000 pixelsUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
vector จากชั้นลึก
Attach timestamps, parameters and code version.
WORKED EXAMPLE
edge_density=5%; compactness=4πA/P²=0.503A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
cnn_embedding__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
histogram O(HW); convolution ≈ O(HWk²CinCout); GPU inference ≈ ms–100ms/imageActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resize, crop, lighting และ camera domain เป็นส่วนหนึ่งของนิยาม Feature ไม่ใช่ preprocessing ที่มองข้ามได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
telemetry · event stream · device Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
temperature=[20,22,23,25]; timestamps every minute; one 5-min gapUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ระดับและความผันผวนใน window
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rolling_mean_std__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/featureActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resampling ต้องระบุ timezone, late event และ gap policy; interpolate อาจสร้างเหตุการณ์ที่ไม่เคยเกิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Lag features
x(t−1), x(t−k)เก็บความจำของระบบ
＋
WHAT IT MEASURES AND WHY
เก็บความจำของระบบ
SOURCE INPUT
telemetry · event stream · device Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
temperature=[20,22,23,25]; timestamps every minute; one 5-min gapUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
x(t−1), x(t−k)
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
lag_features__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/featureActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resampling ต้องระบุ timezone, late event และ gap policy; interpolate อาจสร้างเหตุการณ์ที่ไม่เคยเกิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
03
Rate of change
Δx/Δtเห็นการเปลี่ยนเร็วผิดปกติ
＋
WHAT IT MEASURES AND WHY
เห็นการเปลี่ยนเร็วผิดปกติ
SOURCE INPUT
telemetry · event stream · device Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
temperature=[20,22,23,25]; timestamps every minute; one 5-min gapUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
Δx/Δt
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
rate_of_change__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/featureActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resampling ต้องระบุ timezone, late event และ gap policy; interpolate อาจสร้างเหตุการณ์ที่ไม่เคยเกิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Missingness
gap length, missing ratioคุณภาพ sensor เป็นสัญญาณหนึ่ง
＋
WHAT IT MEASURES AND WHY
คุณภาพ sensor เป็นสัญญาณหนึ่ง
SOURCE INPUT
telemetry · event stream · device Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
temperature=[20,22,23,25]; timestamps every minute; one 5-min gapUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
gap length, missing ratio
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
missingness__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/featureActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resampling ต้องระบุ timezone, late event และ gap policy; interpolate อาจสร้างเหตุการณ์ที่ไม่เคยเกิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Spectral energy
พลังงานตามย่านความถี่จับวงจรที่ time domain มองยาก
＋
WHAT IT MEASURES AND WHY
จับวงจรที่ time domain มองยาก
SOURCE INPUT
telemetry · event stream · device Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
temperature=[20,22,23,25]; timestamps every minute; one 5-min gapUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
พลังงานตามย่านความถี่
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
spectral_energy__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/featureActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
resampling ต้องระบุ timezone, late event และ gap policy; interpolate อาจสร้างเหตุการณ์ที่ไม่เคยเกิด One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DATA EXAMPLE
A small observation
temperature=[20,22,23,25]; timestamps every minute; one 5-min gap
CALCULATION
From data to feature
mean=22.5; slope=(25−20)/3=1.67°C/min; gap_flag=1
COMPUTE TIME
How cost grows
rolling incremental O(1)/event; FFT O(N log N); edge latency µs–ms/feature
coordinate · trajectory · region Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
A=(13.7563,100.5018), B≈1 km east, elapsed=120 secUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ระยะบนทรงกลมระหว่างพิกัด
Attach timestamps, parameters and code version.
WORKED EXAMPLE
distance≈1,000m; speed≈8.33m/s≈30km/hA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
haversine_distance__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
distance O(N); spatial join index ≈ O(N log M); naive pairwise O(NM)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
GPS accuracy, map matching และ privacy มีผลต่อการตีความ; cell boundary ทำให้จุดใกล้กันอยู่คนละกลุ่มได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Speed/acceleration
distance/time และ Δspeed/timeอ่านรูปแบบการเดินทาง
＋
WHAT IT MEASURES AND WHY
อ่านรูปแบบการเดินทาง
SOURCE INPUT
coordinate · trajectory · region Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
A=(13.7563,100.5018), B≈1 km east, elapsed=120 secUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
distance/time และ Δspeed/time
Attach timestamps, parameters and code version.
WORKED EXAMPLE
distance≈1,000m; speed≈8.33m/s≈30km/hA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
speed_acceleration__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
distance O(N); spatial join index ≈ O(N log M); naive pairwise O(NM)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
GPS accuracy, map matching และ privacy มีผลต่อการตีความ; cell boundary ทำให้จุดใกล้กันอยู่คนละกลุ่มได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
coordinate · trajectory · region Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
A=(13.7563,100.5018), B≈1 km east, elapsed=120 secUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
รหัสพื้นที่หลาย resolution
Attach timestamps, parameters and code version.
WORKED EXAMPLE
distance≈1,000m; speed≈8.33m/s≈30km/hA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
geohash_h3_cell__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
distance O(N); spatial join index ≈ O(N log M); naive pairwise O(NM)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
GPS accuracy, map matching และ privacy มีผลต่อการตีความ; cell boundary ทำให้จุดใกล้กันอยู่คนละกลุ่มได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
POI density
จำนวนจุดสนใจใน radiusแทนบริบทของพื้นที่
＋
WHAT IT MEASURES AND WHY
แทนบริบทของพื้นที่
SOURCE INPUT
coordinate · trajectory · region Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
A=(13.7563,100.5018), B≈1 km east, elapsed=120 secUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
จำนวนจุดสนใจใน radius
Attach timestamps, parameters and code version.
WORKED EXAMPLE
distance≈1,000m; speed≈8.33m/s≈30km/hA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
poi_density__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
distance O(N); spatial join index ≈ O(N log M); naive pairwise O(NM)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
GPS accuracy, map matching และ privacy มีผลต่อการตีความ; cell boundary ทำให้จุดใกล้กันอยู่คนละกลุ่มได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Dwell time
เวลาที่อยู่ในพื้นที่แยกผ่านทางออกจากการหยุด
＋
WHAT IT MEASURES AND WHY
แยกผ่านทางออกจากการหยุด
SOURCE INPUT
coordinate · trajectory · region Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
A=(13.7563,100.5018), B≈1 km east, elapsed=120 secUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
เวลาที่อยู่ในพื้นที่
Attach timestamps, parameters and code version.
WORKED EXAMPLE
distance≈1,000m; speed≈8.33m/s≈30km/hA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
dwell_time__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
distance O(N); spatial join index ≈ O(N log M); naive pairwise O(NM)Actual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
GPS accuracy, map matching และ privacy มีผลต่อการตีความ; cell boundary ทำให้จุดใกล้กันอยู่คนละกลุ่มได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
DATA EXAMPLE
A small observation
A=(13.7563,100.5018), B≈1 km east, elapsed=120 sec
activity · assessment · sequence Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
answers=[0,1,1,1]; times=[50,42,35,31] sec; concept=OSI routingUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
correct/attempts แยกทักษะ
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mastery=3/4=75%; time slope=(31−50)/3=−6.33 sec/attemptA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
mastery_estimate__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(events); sequence model O(L)–O(L²); dashboard มัก sub-second หลัง aggregateActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
คะแนนคือหลักฐานบางส่วน ไม่ใช่ตัวตนผู้เรียน; ปกป้องข้อมูลและแยก “ยังไม่เรียน” จาก “ทำไม่ได้” One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Response time
เวลาอ่านและตอบแยกความคล่องจากการเดาได้บางส่วน
＋
WHAT IT MEASURES AND WHY
แยกความคล่องจากการเดาได้บางส่วน
SOURCE INPUT
activity · assessment · sequence Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
answers=[0,1,1,1]; times=[50,42,35,31] sec; concept=OSI routingUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
เวลาอ่านและตอบ
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mastery=3/4=75%; time slope=(31−50)/3=−6.33 sec/attemptA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
response_time__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(events); sequence model O(L)–O(L²); dashboard มัก sub-second หลัง aggregateActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
คะแนนคือหลักฐานบางส่วน ไม่ใช่ตัวตนผู้เรียน; ปกป้องข้อมูลและแยก “ยังไม่เรียน” จาก “ทำไม่ได้” One feature is one piece of evidence—not the complete state of a person, machine, market or system.
activity · assessment · sequence Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
answers=[0,1,1,1]; times=[50,42,35,31] sec; concept=OSI routingUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
คะแนน/ความผิดพลาดตามลำดับ
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mastery=3/4=75%; time slope=(31−50)/3=−6.33 sec/attemptA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
attempt_trajectory__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(events); sequence model O(L)–O(L²); dashboard มัก sub-second หลัง aggregateActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
คะแนนคือหลักฐานบางส่วน ไม่ใช่ตัวตนผู้เรียน; ปกป้องข้อมูลและแยก “ยังไม่เรียน” จาก “ทำไม่ได้” One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Spacing
ช่วงห่างระหว่างการทบทวนอธิบายรูปแบบการฝึก
＋
WHAT IT MEASURES AND WHY
อธิบายรูปแบบการฝึก
SOURCE INPUT
activity · assessment · sequence Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
answers=[0,1,1,1]; times=[50,42,35,31] sec; concept=OSI routingUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ช่วงห่างระหว่างการทบทวน
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mastery=3/4=75%; time slope=(31−50)/3=−6.33 sec/attemptA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
spacing__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(events); sequence model O(L)–O(L²); dashboard มัก sub-second หลัง aggregateActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
คะแนนคือหลักฐานบางส่วน ไม่ใช่ตัวตนผู้เรียน; ปกป้องข้อมูลและแยก “ยังไม่เรียน” จาก “ทำไม่ได้” One feature is one piece of evidence—not the complete state of a person, machine, market or system.
05
Misconception pattern
distribution ของ distractorบอกว่าเข้าใจผิดอย่างไร
＋
WHAT IT MEASURES AND WHY
บอกว่าเข้าใจผิดอย่างไร
SOURCE INPUT
activity · assessment · sequence Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
answers=[0,1,1,1]; times=[50,42,35,31] sec; concept=OSI routingUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
distribution ของ distractor
Attach timestamps, parameters and code version.
WORKED EXAMPLE
mastery=3/4=75%; time slope=(31−50)/3=−6.33 sec/attemptA compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
misconception_pattern__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(events); sequence model O(L)–O(L²); dashboard มัก sub-second หลัง aggregateActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
คะแนนคือหลักฐานบางส่วน ไม่ใช่ตัวตนผู้เรียน; ปกป้องข้อมูลและแยก “ยังไม่เรียน” จาก “ทำไม่ได้” One feature is one piece of evidence—not the complete state of a person, machine, market or system.
Important features—from source observation to a reusable feature contract
Select a feature to open its data, calculation, output contract, computational cost and interpretation boundary.
01
Load factor
average load/peak loadบอกความสม่ำเสมอของการใช้กำลัง
＋
WHAT IT MEASURES AND WHY
บอกความสม่ำเสมอของการใช้กำลัง
SOURCE INPUT
meter · weather · tariff · grid Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
hourly load=[80,100,120,100] MW; renewable=90 MWh; total=400 MWhUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
average load/peak load
Attach timestamps, parameters and code version.
WORKED EXAMPLE
avg=100; peak=120; load_factor=0.833; renewable_share=22.5%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
load_factor__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(T); forecast features across S sites O(ST); streaming ≈ ms/eventActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
อากาศ วันหยุด tariff และ outage ต้องตรงตามเวลาที่รู้จริง; meter revision ต้อง version ได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
02
Ramp rate
ΔMW/Δtimeวัดความเร็วที่ระบบต้องปรับตัว
＋
WHAT IT MEASURES AND WHY
วัดความเร็วที่ระบบต้องปรับตัว
SOURCE INPUT
meter · weather · tariff · grid Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
hourly load=[80,100,120,100] MW; renewable=90 MWh; total=400 MWhUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ΔMW/Δtime
Attach timestamps, parameters and code version.
WORKED EXAMPLE
avg=100; peak=120; load_factor=0.833; renewable_share=22.5%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
ramp_rate__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(T); forecast features across S sites O(ST); streaming ≈ ms/eventActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
อากาศ วันหยุด tariff และ outage ต้องตรงตามเวลาที่รู้จริง; meter revision ต้อง version ได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
03
Degree hours
ความต่างอุณหภูมิจากฐานสะสมแทน demand จากอากาศ
＋
WHAT IT MEASURES AND WHY
แทน demand จากอากาศ
SOURCE INPUT
meter · weather · tariff · grid Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
hourly load=[80,100,120,100] MW; renewable=90 MWh; total=400 MWhUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
ความต่างอุณหภูมิจากฐานสะสม
Attach timestamps, parameters and code version.
WORKED EXAMPLE
avg=100; peak=120; load_factor=0.833; renewable_share=22.5%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
degree_hours__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(T); forecast features across S sites O(ST); streaming ≈ ms/eventActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
อากาศ วันหยุด tariff และ outage ต้องตรงตามเวลาที่รู้จริง; meter revision ต้อง version ได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
04
Peak/off-peak ratio
พลังงานช่วง peak/off-peakอ่านรูปแบบเวลาและ tariff
＋
WHAT IT MEASURES AND WHY
อ่านรูปแบบเวลาและ tariff
SOURCE INPUT
meter · weather · tariff · grid Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
hourly load=[80,100,120,100] MW; renewable=90 MWh; total=400 MWhUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
พลังงานช่วง peak/off-peak
Attach timestamps, parameters and code version.
WORKED EXAMPLE
avg=100; peak=120; load_factor=0.833; renewable_share=22.5%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
peak_off-peak_ratio__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(T); forecast features across S sites O(ST); streaming ≈ ms/eventActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
อากาศ วันหยุด tariff และ outage ต้องตรงตามเวลาที่รู้จริง; meter revision ต้อง version ได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.
meter · weather · tariff · grid Declare entity, source, unit, event time and availability time. A similarly named field from another system may not be equivalent.
SMALL DATA EXAMPLE
hourly load=[80,100,120,100] MW; renewable=90 MWh; total=400 MWhUse only records available at the feature timestamp.
CALCULATION WORKFLOW
Filter the declared entity and time window.
Validate units, missing values and impossible measurements.
renewable/total generation
Attach timestamps, parameters and code version.
WORKED EXAMPLE
avg=100; peak=120; load_factor=0.833; renewable_share=22.5%A compact arithmetic example; production data needs a larger validated reference window.
OUTPUT CONTRACT
renewable_share__value entity_id · as_of · available_at · window · unit · versionThe metadata makes the value reproducible and comparable.
COMPUTATIONAL TIME
aggregation O(T); forecast features across S sites O(ST); streaming ≈ ms/eventActual time depends on volume, window, indexing, hardware, implementation and reuse. Benchmark on the target system.
INTERPRETATION BOUNDARY
อากาศ วันหยุด tariff และ outage ต้องตรงตามเวลาที่รู้จริง; meter revision ต้อง version ได้ One feature is one piece of evidence—not the complete state of a person, machine, market or system.