| 1 | /**
|
| 2 | * @license
|
| 3 | * Copyright 2018 Google LLC
|
| 4 | *
|
| 5 | * Use of this source code is governed by an MIT-style
|
| 6 | * license that can be found in the LICENSE file or at
|
| 7 | * https://opensource.org/licenses/MIT.
|
| 8 | * =============================================================================
|
| 9 | */
|
| 10 | /// <amd-module name="@tensorflow/tfjs-layers/dist/engine/training_utils" />
|
| 11 | import { Tensor } from '@tensorflow/tfjs-core';
|
| 12 | /**
|
| 13 | * For multi-class classification problems, this object is designed to store a
|
| 14 | * mapping from class index to the "weight" of the class, where higher weighted
|
| 15 | * classes have larger impact on loss, accuracy, and other metrics.
|
| 16 | *
|
| 17 | * This is useful for cases in which you want the model to "pay more attention"
|
| 18 | * to examples from an under-represented class, e.g., in unbalanced datasets.
|
| 19 | */
|
| 20 | export declare type ClassWeight = {
|
| 21 | [classIndex: number]: number;
|
| 22 | };
|
| 23 | /**
|
| 24 | * Class weighting for a model with multiple outputs.
|
| 25 | *
|
| 26 | * This object maps each output name to a class-weighting object.
|
| 27 | */
|
| 28 | export declare type ClassWeightMap = {
|
| 29 | [outputName: string]: ClassWeight;
|
| 30 | };
|
| 31 | /**
|
| 32 | * Standardize class weighting objects.
|
| 33 | *
|
| 34 | * This function takes a single class-weighting object, an array of them,
|
| 35 | * or a map from output name to class-weighting object. It compares it to the
|
| 36 | * output name(s) of the model, base on which it outputs an array of
|
| 37 | * class-weighting objects of which the length matches the number of outputs.
|
| 38 | *
|
| 39 | * @param classWeight Input class-weighting object(s).
|
| 40 | * @param outputNames All output name(s) of the model.
|
| 41 | * @return An array of class-weighting objects. The length of the array matches
|
| 42 | * the model's number of outputs.
|
| 43 | */
|
| 44 | export declare function standardizeClassWeights(classWeight: ClassWeight | ClassWeight[] | ClassWeightMap, outputNames: string[]): ClassWeight[];
|
| 45 | export declare function standardizeSampleWeights(classWeight: ClassWeight | ClassWeight[] | ClassWeightMap, outputNames: string[]): ClassWeight[];
|
| 46 | /**
|
| 47 | * Standardize by-sample and/or by-class weights for training.
|
| 48 | *
|
| 49 | * Note that this function operates on one model output at a time. For a model
|
| 50 | * with multiple outputs, you must call this function multiple times.
|
| 51 | *
|
| 52 | * @param y The target tensor that the by-sample and/or by-class weight is for.
|
| 53 | * The values of y are assumed to encode the classes, either directly
|
| 54 | * as an integer index, or as one-hot encoding.
|
| 55 | * @param sampleWeight By-sample weights.
|
| 56 | * @param classWeight By-class weights: an object mapping class indices
|
| 57 | * (integers) to a weight (float) to apply to the model's loss for the
|
| 58 | * samples from this class during training. This can be useful to tell the
|
| 59 | * model to "pay more attention" to samples from an under-represented class.
|
| 60 | * @param sampleWeightMode The mode for the sample weights.
|
| 61 | * @return A Promise of weight tensor, of which the size of the first dimension
|
| 62 | * matches that of `y`.
|
| 63 | */
|
| 64 | export declare function standardizeWeights(y: Tensor, sampleWeight?: Tensor, classWeight?: ClassWeight, sampleWeightMode?: 'temporal'): Promise<Tensor>;
|
| 65 | /**
|
| 66 | * Apply per-sample weights on the loss values from a number of samples.
|
| 67 | *
|
| 68 | * @param losses Loss tensor of shape `[batchSize]`.
|
| 69 | * @param sampleWeights Per-sample weight tensor of shape `[batchSize]`.
|
| 70 | * @returns Tensor of the same shape as`losses`.
|
| 71 | */
|
| 72 | export declare function computeWeightedLoss(losses: Tensor, sampleWeights: Tensor): Tensor<import("/tfjs-core").Rank>;
|