ai.djl.training.util

Class MinMaxScaler

java.lang.Object

ai.djl.training.util.MinMaxScaler

All Implemented Interfaces:

java.lang.AutoCloseable

public class MinMaxScaler
extends java.lang.Object
implements java.lang.AutoCloseable

Transform arrays by scaling each value to a given range. The desired range of transformed data can be set using optRange. the range defaults to 0...1.

After fitting the scaler the fitted values are attached to the same NDManager as the input array.

Constructor Summary

Constructors

Constructor and Description
MinMaxScaler()

Method Summary

Modifier and Type	Method and Description
void	close()
MinMaxScaler	detach() Detaches this MinMaxScaler fitted value from current NDManager's lifecycle.
MinMaxScaler	fit(NDArray data) Computes the minimum and maximum to be used for later scaling.
MinMaxScaler	fit(NDArray data, int[] axises) Computes the minimum and maximum to be used for later scaling.
NDArray	getMax() Returns the value of fittedMax.
NDArray	getMin() Returns the value of fittedMin.
NDArray	inverseTransform(NDArray data) Undoes the transformation of X according to feature_range.
NDArray	inverseTransformi(NDArray data) Undoes the transformation of X according to feature_range as an in-place operation.
MinMaxScaler	optRange(float minRange, float maxRange) Sets desired range of transformed data.
NDArray	transform(NDArray data) Transforms the data using the previous calculated minimum and maximum.
NDArray	transformi(NDArray data) Transforms the data in-place using the previous calculated minimum and maximum.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MinMaxScaler

public MinMaxScaler()

Method Detail

fit

public MinMaxScaler fit(NDArray data,
                        int[] axises)

Computes the minimum and maximum to be used for later scaling.

After fitting the scaler the fitted values are attached to the same NDManager as the input array. reusing the minMaxScaler in the context of other NDManager's is possible by detach() the scaler from the NDManager.

Parameters:

data - used to compute the minimum and maximum used for later scaling

axises - minimum maximum computation along this axises

Returns:

the fitted MinMaxScaler

fit

public MinMaxScaler fit(NDArray data)

Computes the minimum and maximum to be used for later scaling.

After fitting the scaler the fitted values are attached to the same NDManager as the input array. reusing the minMaxScaler in the context of other NDManager's is possible by detach() the scaler from the NDManager.

Parameters:

data - used to compute the minimum and maximum used for later scaling

Returns:

the fitted MinMaxScaler

transform

public NDArray transform(NDArray data)

Transforms the data using the previous calculated minimum and maximum.

if fit() is not executed yet, then the minimum/maximum is computer based on the input data array and used for later computations. X_std = (X - X.min(axis=0)) / (X.max(axis=0) - X.min(axis=0)) X_scaled = X_std * (max - min) + min

Parameters:

data - to get transformed

Returns:

the transformed data, the input array is not changed

transformi

public NDArray transformi(NDArray data)

Transforms the data in-place using the previous calculated minimum and maximum.

if fit() is not called before then the minimum/maximum is computer based on the input data array and used for later computations. X_std = (X - X.min(axis=0)) / (X.max(axis=0) - X.min(axis=0)) X_scaled = X_std * (max - min) + min

Parameters:

data - to get transformed

Returns:

the transformed data (reference to the input data) the input array is changed in-place by this operation

inverseTransform

public NDArray inverseTransform(NDArray data)

Undoes the transformation of X according to feature_range.

Parameters:

data - to get transformed

Returns:

the transformed array

inverseTransformi

public NDArray inverseTransformi(NDArray data)

Undoes the transformation of X according to feature_range as an in-place operation.

Parameters:

data - to get transformed, the data get changed in-place

Returns:

the transformed array

detach

public MinMaxScaler detach()

Detaches this MinMaxScaler fitted value from current NDManager's lifecycle.

this becomes un-managed and it is the user's responsibility to close this. Failure to close the resource might cause your machine to run out of native memory.

After fitting the scaler the fitted values are attached to the same NDManager as the input array.

Re-fitting the scaler after detaching doesn't re-attach the scaler to any NDManager.

Returns:

the detached MinMaxScaler (itself) - to use as a fluent API

optRange

public MinMaxScaler optRange(float minRange,
                             float maxRange)

Sets desired range of transformed data.

Parameters:

minRange - min value for desired range

maxRange - max value for desired range

Returns:

the configured MinMaxScaler

getMin

public NDArray getMin()

Returns the value of fittedMin.

Returns:

the fittedMin value

getMax

public NDArray getMax()

Returns the value of fittedMax.

Returns:

the fittedMax value

close

public void close()

Specified by:

close in interface java.lang.AutoCloseable