AutoFE Operator

Operator List

Basic Operator

info

f represents the numerical features, and c represents the category features.

• AggMin(f, c) : the minimum value of f in each category of feature c.

• AggMax(f, c) : the maximum value of f in each category of characteristic c.

• AggMean(f, c) : the average value of f in characteristic c categories.

• AggMedian(f, c) : the median of f in characteristic c categories.

• AggVar(f, c) : variance of f in characteristic c categories.

• CrossCount([c1, c2, ..]) : based on the count of characteristic list aggregates, the list length is greater than or equal to 2.

• Nunique(c1, c2) : count of unique values of c1 in feature c2 categories.

• Entropy(c) : entropy of each category of feature c.

• Percentile(f) : feature f percentile of each data.

• Combine(c1, c2): character combination of feature c1 and feature c2.

• Count(c) : count of categories of feature c.

• Equal(f1, f2) : judge whether feature f1 and feature f2 are equal.

• Min(f1, f2) : take the smaller value of feature f1 compared to feature f2.

• Max(f1, f2) : take the larger value of feature f1 compared with feature f2.

• Sigmoid(f) : sigmoid nonlinear transformation of feature f.

• Round(f) : rounding feature f.

• Residual(f) : retain the characteristic f to find the number after the decimal point.

• Softmax(f) : gradient logarithmic normalization of discrete probability distribution of finite terms.

• Crossrank(f, c) : The sort of f in each category of feature c.

• Bigger(f1, f2) : Whether the feature f1 is greater than the feature f2.

• Smaller(f1, f2) : Whether the feature f1 is less than the feature f2.

• Log(f) : Perform ${ln}(f)$ operation on the feature f.

• Sin(f) : Perform $sin(f)$ operation on the feature f.

• Cosine(f) : Perform $cos(f)$ operation on the feature f.

Time Series Operator

info

f represents the numerical feature, and w represents the number of windows.

• stddev(f, w) : calculate the standard deviation of feature f in the window.

• ts_max(f, w) : calculates the maximum value of feature f in the window.

• ts_min(f, w) : calculates the minimum value of feature f in the window.

• ts_mean(f, w) : calculate the average value of feature f in the window.

• ts_sum(f, w) : calculates the summation of feature f in the window.

• ts_rank(f, w) : calculates the ranking of the current value of feature f in the window (descending).

• ts_argmax(f, w) : calculates the location index of the maximum value of feature f in the window (counted from 0).

• ts_argmin(f, w) : calculates the location index of the minimum value of feature f in the window (counted from 0).

• delay(f, w) : gets the value corresponding to the earliest time of feature f in the window.

• decay(f, w) : calculate the linear attenuation sum of characteristic f in the window.。

• delta(f, w) : calculates the difference between the latest and earliest values of feature f in the window.

• correlation(f1, f2, w) : calculate the correlation coefficient between feature f1 and feature f2 in the window.

• covariance(f1, f2, w) : calculate the covariance of feature f1 and feature f2 in the window.

Operator Diagram

AggMin(f, c)

info

Represents the minimum value of f in all categories of feature c.

AggMax(f, c)

info

Represents the maximum value of f in all categories of feature c.

AggMean(f, c)

info

Represents the average value of f in all categories of feature c.

AggMedian(f, c)

info

Represents the median of f in all categories of feature c.

AggVar(f, c)

info

Represents the variance of f in each category of characteristic c.

CrossCount([c1, c2, ..])

info

According to the count of characteristic list aggregates, the list length is greater than or equal to 2.

Nunique(c1, c2)

info

Represents the count of unique values of c1 in feature c2 categories.

Entropy(c)

info

Represents the entropy of each category of feature c.

Percentile(f)

info

Represents the percentile of each data of feature f.

Combine(c1, c2)

info

Represents the character combination of feature c1 and feature c2.

Count(c)

info

Count of categories of feature c.

Equal(f1, f2)

info

Judge whether feature f1 and feature f2 are equal.

Min(f1, f2)

info

Take the smaller value of feature f1 compared with feature f2.

Max(f1, f2)

info

Take the larger value of feature f1 compared with feature f2.

Sigmoid(f)

info

Sigmoid nonlinear transformation of feature f.

Round(f)

info

Rounding feature f.

Residual(f)

info

Retain the characteristic f to calculate the value after the decimal point.

Softmax(f)

info

Gradient logarithmic normalization of discrete probability distribution of finite terms.

stddev(f, w)

info

Calculate the standard deviation of feature f in the window.

ts_max(f, w)

info

Calculate the maximum value of feature f in the window.

ts_min(f, w)

info

Calculate the minimum value of feature f in the window.

ts_mean(f, w)

info

Calculate the average value of feature f in the window.

ts_sum(f, w)

info

Calculate the sum of feature f in the window.

ts_rank(f, w)

info

Calculate the ranking of the current value of feature f in the window (descending).

ts_argmax(f, w)

info

Calculate the location index of the maximum value of feature f in the window (counted from 0).

ts_argmin(f, w)

info

Calculate the minimum location index of feature f in the window (count from 0).

delay(f, w)

info

Get the value corresponding to the earliest time of feature f in the window.

decay(f, w)

info

Calculate the linear attenuation sum of characteristic f in the window.

delta(f, w)

info

Calculate the difference between the latest and earliest values of feature f in the window.

correlation(f1, f2, w)

info

Calculate the correlation coefficient between feature f1 and feature f2 in the window.

covariance(f1, f2, w)

info

Calculate the covariance of feature f1 and feature f2 in the window.