&Vf ddlZddlmZddlmZddlmZddlmZddlmZddl m Z ddl m Z Gd d eZ ed Gd d e ZedGdde ZedGdde ZedGdde ZedGdde ZedGdde ZedGdde Zed Gd!d"e Zed#Gd$d%e Zed&Gd'd(e Zed)Gd*d+e Zed,Gd-d.e Zed/Gd0d1e Zed2Gd3d4e Zed5Gd6d7e Zed8Gd9d:e Zed;Gd<d=e Zd>Zed?d@gdAZ edBdCgdDZ!edEdFgdGZ"egdHdIZ#egdJdKZ$egdLdMZ%egdNdOZ&edPddRZ'edSdTgddVZ(egdWdXZ)egdYdZZ*ed[d\gd]Z+ed^d_g ddbZ,edcddg ddgZ-edhdig ddjZ.edkdlg ddmZ/edndog ddpZ0edqGdrdse Z1edtduZ2edvGdwdxe Z3edydzZ4ed{Gd|d}e Z5ed~ddZ6dS)N)backend)ops) keras_export)Loss)squeeze_or_expand_to_same_rank)serialization_lib) normalizecHeZdZ dfd ZdZfdZedZxZS)LossFunctionWrappersum_over_batch_sizeNc jt||||_||_dSN) reductionname)super__init__fn _fn_kwargs)selfrrrkwargs __class__s T/var/www/html/software/conda/lib/python3.11/site-packages/keras/src/losses/losses.pyrzLossFunctionWrapper.__init__ s4 94888 cPt||\}}|j||fi|jSN)rrr)ry_truey_preds rcallzLossFunctionWrapper.calls37GGtwvv99999rct}dtj|ji}|tj|ji||S)Nr)r get_configrserialize_keras_objectrupdater)r base_configconfigrs rr zLossFunctionWrapper.get_configs`gg((** )@IIJ '>tOOPPP(+(((rcBd|vrtj|}|di|S)Nr)rdeserialize_keras_object)clsr$s r from_configzLossFunctionWrapper.from_configs- 6>>&?GGFs}}V}}r)r N) __name__ __module__ __qualname__rrr classmethodr) __classcell__rs@rr r s8<!!!!!!:::))))) [rr zkeras.losses.MeanSquaredErrorc,eZdZdZ dfd ZdZxZS)MeanSquaredErroraComputes the mean of squares of errors between labels and predictions. Formula: ```python loss = mean(square(y_true - y_pred)) ``` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r mean_squared_errorcZtt||dSr)rrr2rrrrs rrzMeanSquaredError.__init__6s+ +ytLLLLLrc*tj|Srrr rs rr zMeanSquaredError.get_config;t$$$r)r r2r*r+r,__doc__rr r.r/s@rr1r1%sb   5IMMMMMM %%%%%%%rr1zkeras.losses.MeanAbsoluteErrorc,eZdZdZ dfd ZdZxZS)MeanAbsoluteErroraComputes the mean of absolute difference between labels and predictions. Formula: ```python loss = mean(abs(y_true - y_pred)) ``` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r mean_absolute_errorcZtt||dSr)rrr=r4s rrzMeanAbsoluteError.__init__Ps+ , MMMMMrc*tj|Srr6r7s rr zMeanAbsoluteError.get_configUr8r)r r=r9r/s@rr<r<?sb   5JNNNNNN %%%%%%%rr<z(keras.losses.MeanAbsolutePercentageErrorc.eZdZdZ dfd ZdZxZS)MeanAbsolutePercentageErroraComputes the mean absolute percentage error between `y_true` & `y_pred`. Formula: ```python loss = 100 * mean(abs((y_true - y_pred) / y_true)) ``` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r mean_absolute_percentage_errorcZtt||dSr)rrrBr4s rrz$MeanAbsolutePercentageError.__init__j7  *id      rc*tj|Srr6r7s rr z&MeanAbsolutePercentageError.get_configsr8r)r rBr9r/s@rrArAY^  "( -      %%%%%%%rrAz(keras.losses.MeanSquaredLogarithmicErrorc.eZdZdZ dfd ZdZxZS)MeanSquaredLogarithmicErroraComputes the mean squared logarithmic error between `y_true` & `y_pred`. Formula: ```python loss = mean(square(log(y_true + 1) - log(y_pred + 1))) ``` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r mean_squared_logarithmic_errorcZtt||dSr)rrrIr4s rrz$MeanSquaredLogarithmicError.__init__rDrc*tj|Srr6r7s rr z&MeanSquaredLogarithmicError.get_configr8r)r rIr9r/s@rrHrHwrFrrHzkeras.losses.CosineSimilarityc0eZdZdZ dfd ZdZxZS)CosineSimilarityaComputes the cosine similarity between `y_true` & `y_pred`. Note that it is a number between -1 and 1. When it is a negative number between -1 and 0, 0 indicates orthogonality and values closer to -1 indicate greater similarity. This makes it usable as a loss function in a setting where you try to maximize the proximity between predictions and targets. If either `y_true` or `y_pred` is a zero vector, cosine similarity will be 0 regardless of the proximity between predictions and targets. Formula: ```python loss = -sum(l2_norm(y_true) * l2_norm(y_pred)) ``` Args: axis: The axis along which the cosine similarity is computed (the features axis). Defaults to `-1`. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r cosine_similarityc\tt|||dS)N)rraxis)rrrO)rrQrrrs rrzCosineSimilarity.__init__s9  D      rc*tj|Srr6r7s rr zCosineSimilarity.get_configr8r)rNr rOr9r/s@rrMrMsa4'       %%%%%%%rrMzkeras.losses.Huberc0eZdZdZ dfd ZdZxZS)HuberaComputes the Huber loss between `y_true` & `y_pred`. Formula: ```python for x in error: if abs(x) <= delta: loss.append(0.5 * x^2) elif abs(x) > delta: loss.append(delta * abs(x) - 0.5 * delta^2) loss = mean(loss, axis=-1) ``` See: [Huber loss](https://en.wikipedia.org/wiki/Huber_loss). Args: delta: A float, the point where the Huber loss function changes from a quadratic to linear. reduction: Type of reduction to apply to loss. Options are `"sum"`, `"sum_over_batch_size"` or `None`. Defaults to `"sum_over_batch_size"`. name: Optional name for the instance. ?r huber_lossc\tt|||dS)N)rrdelta)rrhuber)rrXrrrs rrzHuber.__init__s, TYeLLLLLrc*tj|Srr6r7s rr zHuber.get_configr8r)rUr rVr9r/s@rrTrTsg4'  MMMMMM%%%%%%%rrTzkeras.losses.LogCoshc*eZdZdZdfd ZdZxZS)LogCoshaComputes the logarithm of the hyperbolic cosine of the prediction error. Formula: ```python error = y_pred - y_true logcosh = mean(log((exp(error) + exp(-error))/2), axis=-1)` ``` where x is the error `y_pred - y_true`. Args: reduction: Type of reduction to apply to loss. Options are `"sum"`, `"sum_over_batch_size"` or `None`. Defaults to `"sum_over_batch_size"`. name: Optional name for the instance. r log_coshcZtt||dSNrr)rrr]r4s rrzLogCosh.__init__s(  BBBBBrc*tj|Srr6r7s rr zLogCosh.get_configr8r)r r]r9r/s@rr\r\s\"CCCCCC%%%%%%%rr\zkeras.losses.Hingec*eZdZdZdfd ZdZxZS)HingeaComputes the hinge loss between `y_true` & `y_pred`. Formula: ```python loss = maximum(1 - y_true * y_pred, 0) ``` `y_true` values are expected to be -1 or 1. If binary (0 or 1) labels are provided we will convert them to -1 or 1. Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r hingecZtt||dSr)rrrdr4s rrzHinge.__init__s( )$?????rc*tj|Srr6r7s rr zHinge.get_configr8r)r rdr9r/s@rrcrcs\$@@@@@@%%%%%%%rrczkeras.losses.SquaredHingec*eZdZdZdfd ZdZxZS) SquaredHingea%Computes the squared hinge loss between `y_true` & `y_pred`. Formula: ```python loss = square(maximum(1 - y_true * y_pred, 0)) ``` `y_true` values are expected to be -1 or 1. If binary (0 or 1) labels are provided we will convert them to -1 or 1. Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r squared_hingecZtt||dSr)rrrir4s rrzSquaredHinge.__init__,( )$GGGGGrc*tj|Srr6r7s rr zSquaredHinge.get_config/r8r)r rir9r/s@rrhrhs\$HHHHHH%%%%%%%rrhzkeras.losses.CategoricalHingec,eZdZdZ dfd ZdZxZS)CategoricalHingeaComputes the categorical hinge loss between `y_true` & `y_pred`. Formula: ```python loss = maximum(neg - pos + 1, 0) ``` where `neg=maximum((1-y_true)*y_pred)` and `pos=sum(y_true*y_pred)` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r categorical_hingecZtt||dSr)rrror4s rrzCategoricalHinge.__init__Fs+ *idKKKKKrc*tj|Srr6r7s rr zCategoricalHinge.get_configKr8r)r ror9r/s@rrnrn3sb$5HLLLLLL %%%%%%%rrnzkeras.losses.KLDivergencec*eZdZdZdfd ZdZxZS) KLDivergenceaCComputes Kullback-Leibler divergence loss between `y_true` & `y_pred`. Formula: ```python loss = y_true * log(y_true / y_pred) ``` `y_true` and `y_pred` are expected to be probability distributions, with values between 0 and 1. They will get clipped to the `[0, 1]` range. Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r kl_divergencecZtt||dSr)rrrtr4s rrzKLDivergence.__init__drkrc*tj|Srr6r7s rr zKLDivergence.get_configgr8r)r rtr9r/s@rrsrsOs\&HHHHHH%%%%%%%rrszkeras.losses.Poissonc*eZdZdZdfd ZdZxZS)PoissonaComputes the Poisson loss between `y_true` & `y_pred`. Formula: ```python loss = y_pred - y_true * log(y_pred) ``` Args: reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. r poissoncZtt||dSr)rrryr4s rrzPoisson.__init__|s( IDAAAAArc*tj|Srr6r7s rr zPoisson.get_configr8r)r ryr9r/s@rrxrxks\  BBBBBB%%%%%%%rrxzkeras.losses.BinaryCrossentropyc4eZdZdZ d fd ZdZxZS) BinaryCrossentropya{ Computes the cross-entropy loss between true labels and predicted labels. Use this cross-entropy loss for binary (0 or 1) classification applications. The loss function requires the following inputs: - `y_true` (true label): This is either 0 or 1. - `y_pred` (predicted value): This is the model's prediction, i.e, a single floating-point value which either represents a [logit](https://en.wikipedia.org/wiki/Logit), (i.e, value in [-inf, inf] when `from_logits=True`) or a probability (i.e, value in [0., 1.] when `from_logits=False`). Args: from_logits: Whether to interpret `y_pred` as a tensor of [logit](https://en.wikipedia.org/wiki/Logit) values. By default, we assume that `y_pred` is probabilities (i.e., values in [0, 1]). label_smoothing: Float in range [0, 1]. When 0, no smoothing occurs. When > 0, we compute the loss between the predicted labels and a smoothed version of the true labels, where the smoothing squeezes the labels towards 0.5. Larger values of `label_smoothing` correspond to heavier smoothing. axis: The axis along which to compute crossentropy (the features axis). Defaults to `-1`. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. Examples: **Recommended Usage:** (set `from_logits=True`) With `compile()` API: ```python model.compile( loss=keras.losses.BinaryCrossentropy(from_logits=True), ... ) ``` As a standalone function: >>> # Example 1: (batch_size = 1, number of samples = 4) >>> y_true = [0, 1, 0, 0] >>> y_pred = [-18.6, 0.51, 2.94, -12.8] >>> bce = keras.losses.BinaryCrossentropy(from_logits=True) >>> bce(y_true, y_pred) 0.865 >>> # Example 2: (batch_size = 2, number of samples = 4) >>> y_true = [[0, 1], [0, 0]] >>> y_pred = [[-18.6, 0.51], [2.94, -12.8]] >>> # Using default 'auto'/'sum_over_batch_size' reduction type. >>> bce = keras.losses.BinaryCrossentropy(from_logits=True) >>> bce(y_true, y_pred) 0.865 >>> # Using 'sample_weight' attribute >>> bce(y_true, y_pred, sample_weight=[0.8, 0.2]) 0.243 >>> # Using 'sum' reduction` type. >>> bce = keras.losses.BinaryCrossentropy(from_logits=True, ... reduction="sum") >>> bce(y_true, y_pred) 1.730 >>> # Using 'none' reduction type. >>> bce = keras.losses.BinaryCrossentropy(from_logits=True, ... reduction=None) >>> bce(y_true, y_pred) array([0.235, 1.496], dtype=float32) **Default Usage:** (set `from_logits=False`) >>> # Make the following updates to the above "Recommended Usage" section >>> # 1. Set `from_logits=False` >>> keras.losses.BinaryCrossentropy() # OR ...('from_logits=False') >>> # 2. Update `y_pred` to use probabilities instead of logits >>> y_pred = [0.6, 0.3, 0.2, 0.8] # OR [[0.6, 0.3], [0.2, 0.8]] FrNr binary_crossentropyctt|||||||_||_||_dSNrr from_logitslabel_smoothingrQ)rrrrrrQrrrrQrrrs rrzBinaryCrossentropy.__init__sU  #+    '. rcD|j|j|j|j|jdSrrr7s rr zBinaryCrossentropy.get_config,I+#3I    r)Fr~rNr rr9r/s@rr}r}sjNNd ' " (       rr}z$keras.losses.BinaryFocalCrossentropyc:eZdZdZ d fd Zd ZxZS) BinaryFocalCrossentropyaComputes focal cross-entropy loss between true labels and predictions. Binary cross-entropy loss is often used for binary (0 or 1) classification tasks. The loss function requires the following inputs: - `y_true` (true label): This is either 0 or 1. - `y_pred` (predicted value): This is the model's prediction, i.e, a single floating-point value which either represents a [logit](https://en.wikipedia.org/wiki/Logit), (i.e, value in [-inf, inf] when `from_logits=True`) or a probability (i.e, value in `[0., 1.]` when `from_logits=False`). According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it helps to apply a "focal factor" to down-weight easy examples and focus more on hard examples. By default, the focal tensor is computed as follows: `focal_factor = (1 - output) ** gamma` for class 1 `focal_factor = output ** gamma` for class 0 where `gamma` is a focusing parameter. When `gamma=0`, this function is equivalent to the binary crossentropy loss. Args: apply_class_balancing: A bool, whether to apply weight balancing on the binary classes 0 and 1. alpha: A weight balancing factor for class 1, default is `0.25` as mentioned in reference [Lin et al., 2018]( https://arxiv.org/pdf/1708.02002.pdf). The weight for class 0 is `1.0 - alpha`. gamma: A focusing parameter used to compute the focal factor, default is `2.0` as mentioned in the reference [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf). from_logits: Whether to interpret `y_pred` as a tensor of [logit](https://en.wikipedia.org/wiki/Logit) values. By default, we assume that `y_pred` are probabilities (i.e., values in `[0, 1]`). label_smoothing: Float in `[0, 1]`. When `0`, no smoothing occurs. When > `0`, we compute the loss between the predicted labels and a smoothed version of the true labels, where the smoothing squeezes the labels towards `0.5`. Larger values of `label_smoothing` correspond to heavier smoothing. axis: The axis along which to compute crossentropy (the features axis). Defaults to `-1`. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. Examples: With the `compile()` API: ```python model.compile( loss=keras.losses.BinaryFocalCrossentropy( gamma=2.0, from_logits=True), ... ) ``` As a standalone function: >>> # Example 1: (batch_size = 1, number of samples = 4) >>> y_true = [0, 1, 0, 0] >>> y_pred = [-18.6, 0.51, 2.94, -12.8] >>> loss = keras.losses.BinaryFocalCrossentropy( ... gamma=2, from_logits=True) >>> loss(y_true, y_pred) 0.691 >>> # Apply class weight >>> loss = keras.losses.BinaryFocalCrossentropy( ... apply_class_balancing=True, gamma=2, from_logits=True) >>> loss(y_true, y_pred) 0.51 >>> # Example 2: (batch_size = 2, number of samples = 4) >>> y_true = [[0, 1], [0, 0]] >>> y_pred = [[-18.6, 0.51], [2.94, -12.8]] >>> # Using default 'auto'/'sum_over_batch_size' reduction type. >>> loss = keras.losses.BinaryFocalCrossentropy( ... gamma=3, from_logits=True) >>> loss(y_true, y_pred) 0.647 >>> # Apply class weight >>> loss = keras.losses.BinaryFocalCrossentropy( ... apply_class_balancing=True, gamma=3, from_logits=True) >>> loss(y_true, y_pred) 0.482 >>> # Using 'sample_weight' attribute with focal effect >>> loss = keras.losses.BinaryFocalCrossentropy( ... gamma=3, from_logits=True) >>> loss(y_true, y_pred, sample_weight=[0.8, 0.2]) 0.133 >>> # Apply class weight >>> loss = keras.losses.BinaryFocalCrossentropy( ... apply_class_balancing=True, gamma=3, from_logits=True) >>> loss(y_true, y_pred, sample_weight=[0.8, 0.2]) 0.097 >>> # Using 'sum' reduction` type. >>> loss = keras.losses.BinaryFocalCrossentropy( ... gamma=4, from_logits=True, ... reduction="sum") >>> loss(y_true, y_pred) 1.222 >>> # Apply class weight >>> loss = keras.losses.BinaryFocalCrossentropy( ... apply_class_balancing=True, gamma=4, from_logits=True, ... reduction="sum") >>> loss(y_true, y_pred) 0.914 >>> # Using 'none' reduction type. >>> loss = keras.losses.BinaryFocalCrossentropy( ... gamma=5, from_logits=True, ... reduction=None) >>> loss(y_true, y_pred) array([0.0017 1.1561], dtype=float32) >>> # Apply class weight >>> loss = keras.losses.BinaryFocalCrossentropy( ... apply_class_balancing=True, gamma=5, from_logits=True, ... reduction=None) >>> loss(y_true, y_pred) array([0.0004 0.8670], dtype=float32) F?@r~rNr binary_focal_crossentropyc tt|||||||| ||_||_||_||_||_||_dS)N)apply_class_balancingalphagammarrrrrQ) rrrrrrQrrr) rrrrrrrQrrrs rrz BinaryFocalCrossentropy.__init__wst  %"7#+  '. %:"  rc h|j|j|j|j|j|j|j|jdS)NrrrrrQrrrrr7s rr z"BinaryFocalCrossentropy.get_configs<I+#3I%)%?ZZ   r)FrrFr~rNr rr9r/s@rrrss@@H$ ' (:        rrz$keras.losses.CategoricalCrossentropyc4eZdZdZ d fd ZdZxZS) CategoricalCrossentropyaComputes the crossentropy loss between the labels and predictions. Use this crossentropy loss function when there are two or more label classes. We expect labels to be provided in a `one_hot` representation. If you want to provide labels as integers, please use `SparseCategoricalCrossentropy` loss. There should be `num_classes` floating point values per feature, i.e., the shape of both `y_pred` and `y_true` are `[batch_size, num_classes]`. Args: from_logits: Whether `y_pred` is expected to be a logits tensor. By default, we assume that `y_pred` encodes a probability distribution. label_smoothing: Float in [0, 1]. When > 0, label values are smoothed, meaning the confidence on label values are relaxed. For example, if `0.1`, use `0.1 / num_classes` for non-target labels and `0.9 + 0.1 / num_classes` for target labels. axis: The axis along which to compute crossentropy (the features axis). Defaults to `-1`. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. Examples: Standalone usage: >>> y_true = [[0, 1, 0], [0, 0, 1]] >>> y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]] >>> # Using 'auto'/'sum_over_batch_size' reduction type. >>> cce = keras.losses.CategoricalCrossentropy() >>> cce(y_true, y_pred) 1.177 >>> # Calling with 'sample_weight'. >>> cce(y_true, y_pred, sample_weight=np.array([0.3, 0.7])) 0.814 >>> # Using 'sum' reduction type. >>> cce = keras.losses.CategoricalCrossentropy( ... reduction="sum") >>> cce(y_true, y_pred) 2.354 >>> # Using 'none' reduction type. >>> cce = keras.losses.CategoricalCrossentropy( ... reduction=None) >>> cce(y_true, y_pred) array([0.0513, 2.303], dtype=float32) Usage with the `compile()` API: ```python model.compile(optimizer='sgd', loss=keras.losses.CategoricalCrossentropy()) ``` Fr~rNr categorical_crossentropyctt|||||||_||_||_dSr)rrrrrrQrs rrz CategoricalCrossentropy.__init__sU  $#+    '. rcD|j|j|j|j|jdSrrr7s rr z"CategoricalCrossentropy.get_configrr)Fr~rNr rr9r/s@rrrsh88x ' ' (       rrz)keras.losses.CategoricalFocalCrossentropyc8eZdZdZ d fd Zd ZxZS) CategoricalFocalCrossentropya/Computes the alpha balanced focal crossentropy loss. Use this crossentropy loss function when there are two or more label classes and if you want to handle class imbalance without using `class_weights`. We expect labels to be provided in a `one_hot` representation. According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it helps to apply a focal factor to down-weight easy examples and focus more on hard examples. The general formula for the focal loss (FL) is as follows: `FL(p_t) = (1 - p_t) ** gamma * log(p_t)` where `p_t` is defined as follows: `p_t = output if y_true == 1, else 1 - output` `(1 - p_t) ** gamma` is the `modulating_factor`, where `gamma` is a focusing parameter. When `gamma` = 0, there is no focal effect on the cross entropy. `gamma` reduces the importance given to simple examples in a smooth manner. The authors use alpha-balanced variant of focal loss (FL) in the paper: `FL(p_t) = -alpha * (1 - p_t) ** gamma * log(p_t)` where `alpha` is the weight factor for the classes. If `alpha` = 1, the loss won't be able to handle class imbalance properly as all classes will have the same weight. This can be a constant or a list of constants. If alpha is a list, it must have the same length as the number of classes. The formula above can be generalized to: `FL(p_t) = alpha * (1 - p_t) ** gamma * CrossEntropy(y_true, y_pred)` where minus comes from `CrossEntropy(y_true, y_pred)` (CE). Extending this to multi-class case is straightforward: `FL(p_t) = alpha * (1 - p_t) ** gamma * CategoricalCE(y_true, y_pred)` In the snippet below, there is `num_classes` floating pointing values per example. The shape of both `y_pred` and `y_true` are `(batch_size, num_classes)`. Args: alpha: A weight balancing factor for all classes, default is `0.25` as mentioned in the reference. It can be a list of floats or a scalar. In the multi-class case, alpha may be set by inverse class frequency by using `compute_class_weight` from `sklearn.utils`. gamma: A focusing parameter, default is `2.0` as mentioned in the reference. It helps to gradually reduce the importance given to simple (easy) examples in a smooth manner. from_logits: Whether `output` is expected to be a logits tensor. By default, we consider that `output` encodes a probability distribution. label_smoothing: Float in [0, 1]. When > 0, label values are smoothed, meaning the confidence on label values are relaxed. For example, if `0.1`, use `0.1 / num_classes` for non-target labels and `0.9 + 0.1 / num_classes` for target labels. axis: The axis along which to compute crossentropy (the features axis). Defaults to `-1`. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. Examples: Standalone usage: >>> y_true = [[0., 1., 0.], [0., 0., 1.]] >>> y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]] >>> # Using 'auto'/'sum_over_batch_size' reduction type. >>> cce = keras.losses.CategoricalFocalCrossentropy() >>> cce(y_true, y_pred) 0.23315276 >>> # Calling with 'sample_weight'. >>> cce(y_true, y_pred, sample_weight=np.array([0.3, 0.7])) 0.1632 >>> # Using 'sum' reduction type. >>> cce = keras.losses.CategoricalFocalCrossentropy( ... reduction="sum") >>> cce(y_true, y_pred) 0.46631 >>> # Using 'none' reduction type. >>> cce = keras.losses.CategoricalFocalCrossentropy( ... reduction=None) >>> cce(y_true, y_pred) array([3.2058331e-05, 4.6627346e-01], dtype=float32) Usage with the `compile()` API: ```python model.compile(optimizer='adam', loss=keras.losses.CategoricalFocalCrossentropy()) ``` rrFr~rNr categorical_focal_crossentropyc tt|||||||||_||_||_||_||_dS)z4Initializes `CategoricalFocalCrossentropy` instance.)rrrrrrrQN)rrrrrrQrr) rrrrrrQrrrs rrz%CategoricalFocalCrossentropy.__init__`si  *#+  '.   rc\|j|j|j|j|j|j|jdS)NrrrrrQrrrr7s rr z'CategoricalFocalCrossentropy.get_config{s6I+#3IZZ   r)rrFr~rNr rr9r/s@rrrspaaJ ' -6        rrz*keras.losses.SparseCategoricalCrossentropyc2eZdZdZ dfd ZdZxZS) SparseCategoricalCrossentropyaComputes the crossentropy loss between the labels and predictions. Use this crossentropy loss function when there are two or more label classes. We expect labels to be provided as integers. If you want to provide labels using `one-hot` representation, please use `CategoricalCrossentropy` loss. There should be `# classes` floating point values per feature for `y_pred` and a single floating point value per feature for `y_true`. In the snippet below, there is a single floating point value per example for `y_true` and `num_classes` floating pointing values per example for `y_pred`. The shape of `y_true` is `[batch_size]` and the shape of `y_pred` is `[batch_size, num_classes]`. Args: from_logits: Whether `y_pred` is expected to be a logits tensor. By default, we assume that `y_pred` encodes a probability distribution. reduction: Type of reduction to apply to the loss. In almost all cases this should be `"sum_over_batch_size"`. Supported options are `"sum"`, `"sum_over_batch_size"` or `None`. name: Optional name for the loss instance. Examples: >>> y_true = [1, 2] >>> y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]] >>> # Using 'auto'/'sum_over_batch_size' reduction type. >>> scce = keras.losses.SparseCategoricalCrossentropy() >>> scce(y_true, y_pred) 1.177 >>> # Calling with 'sample_weight'. >>> scce(y_true, y_pred, sample_weight=np.array([0.3, 0.7])) 0.814 >>> # Using 'sum' reduction type. >>> scce = keras.losses.SparseCategoricalCrossentropy( ... reduction="sum") >>> scce(y_true, y_pred) 2.354 >>> # Using 'none' reduction type. >>> scce = keras.losses.SparseCategoricalCrossentropy( ... reduction=None) >>> scce(y_true, y_pred) array([0.0513, 2.303], dtype=float32) Usage with the `compile()` API: ```python model.compile(optimizer='sgd', loss=keras.losses.SparseCategoricalCrossentropy()) ``` FNr sparse_categorical_crossentropycztt||||||_||_dSNrrr ignore_class)rrrrr)rrrrrrs rrz&SparseCategoricalCrossentropy.__init__sM  +#%    '(rc8|j|j|j|jdSrrr7s rr z(SparseCategoricalCrossentropy.get_configs'I+ -    r)FNr rr9r/s@rrrse55r' . ))))))"       rrctjd}tjd}tjtj||}fd}fd}tj|||}|S)zCConverts binary labels into -1/1 for hinge loss/metric calculation.rcdzdz S)NrrUr&rsr_convert_binary_labelsz>convert_binary_labels_to_hinge.._convert_binary_labelssV|c!!rcSrr&rsr_return_labels_unconvertedzBconvert_binary_labels_to_hinge.._return_labels_unconverteds r)requalall logical_orcond)r are_zerosare_ones is_binaryrrupdated_y_trues` rconvert_binary_labels_to_hingers &!$$Iy##H 8<<>>I"""""X)+EN rzkeras.metrics.hingezkeras.losses.hingectj|}tj||j}tj|}t |}tjtjd||zz ddS)aComputes the hinge loss between `y_true` & `y_pred`. Formula: ```python loss = mean(maximum(1 - y_true * y_pred, 0), axis=-1) ``` Args: y_true: The ground truth values. `y_true` values are expected to be -1 or 1. If binary (0 or 1) labels are provided they will be converted to -1 or 1 with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Hinge loss values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.choice([-1, 1], size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.hinge(y_true, y_pred) dtyperUr~rNrQ)rconvert_to_tensorcastrrmeanmaximumrrs rrdrdsq< "6 * *F XfFL 1 1 1F  "6 * *F +F 3 3F 8CKfvo 5s;;" E E EErzkeras.metrics.squared_hingezkeras.losses.squared_hingec tj|}tj||j}t |}tjtjtjd||zz ddS)aComputes the squared hinge loss between `y_true` & `y_pred`. Formula: ```python loss = mean(square(maximum(1 - y_true * y_pred, 0)), axis=-1) ``` Args: y_true: The ground truth values. `y_true` values are expected to be -1 or 1. If binary (0 or 1) labels are provided we will convert them to -1 or 1 with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Squared hinge loss values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.choice([-1, 1], size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.squared_hinge(y_true, y_pred) rUr~rNr)rrrrrrsquarerrs rririsq< "6 * *F Xffl + +F +F 3 3F 8 3;sVf_4c::;;"   rzkeras.metrics.categorical_hingezkeras.losses.categorical_hingec2tj|}tj||j}tj||zd}tjd|z |zd}tjd|j}tj||z dz|S)a<Computes the categorical hinge loss between `y_true` & `y_pred`. Formula: ```python loss = maximum(neg - pos + 1, 0) ``` where `neg=maximum((1-y_true)*y_pred)` and `pos=sum(y_true*y_pred)` Args: y_true: The ground truth values. `y_true` values are expected to be either `{-1, +1}` or `{0, 1}` (i.e. a one-hot-encoded tensor) with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Categorical hinge loss values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.randint(0, 3, size=(2,)) >>> y_true = np.eye(np.max(y_true) + 1)[y_true] >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.categorical_hinge(y_true, y_pred) rNrrUr~)rrrrsummaxr)rrposnegzeros rroro9sB "6 * *F Xffl + +F '&6/ + + +C '3<6) 3 3 3C 8C & &D ;sSy3 - --r)z keras.metrics.mean_squared_errorzkeras.losses.mean_squared_errorzkeras._legacy.losses.msezkeras._legacy.losses.MSEzkeras._legacy.metrics.msezkeras._legacy.metrics.MSEctj|}tj||j}t||\}}tjtj||z dS)aLComputes the mean squared error between labels and predictions. Formula: ```python loss = mean(square(y_true - y_pred), axis=-1) ``` Example: >>> y_true = np.random.randint(0, 2, size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.mean_squared_error(y_true, y_pred) Args: y_true: Ground truth values with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Mean squared error values with shape = `[batch_size, d0, .. dN-1]`. rrNr)rrrrrrrs rr2r2bsbB "6 * *F  "6 > > >F3FFCCNFF 8CJv//b 9 9 99r)z!keras.metrics.mean_absolute_errorz keras.losses.mean_absolute_errorzkeras._legacy.losses.MAEzkeras._legacy.losses.maezkeras._legacy.metrics.MAEzkeras._legacy.metrics.maectj|}tj||j}t||\}}tjtj||z dS)a>Computes the mean absolute error between labels and predictions. ```python loss = mean(abs(y_true - y_pred), axis=-1) ``` Args: y_true: Ground truth values with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Mean absolute error values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.randint(0, 2, size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.mean_absolute_error(y_true, y_pred) rrNr)rrrrrabsrs rr=r=sa> "6 * *F  "6 > > >F3FFCCNFF 8CGFVO,,2 6 6 66r)z,keras.metrics.mean_absolute_percentage_errorz+keras.losses.mean_absolute_percentage_errorzkeras._legacy.losses.mapezkeras._legacy.losses.MAPEzkeras._legacy.metrics.mapezkeras._legacy.metrics.MAPEctj|}tj||j}tjtj}t ||\}}tj||z tjtj||z }dtj|dzS)aComputes the mean absolute percentage error between `y_true` & `y_pred`. Formula: ```python loss = 100 * mean(abs((y_true - y_pred) / y_true), axis=-1) ``` Division by zero is prevented by dividing by `maximum(y_true, epsilon)` where `epsilon = keras.backend.epsilon()` (default to `1e-7`). Args: y_true: Ground truth values with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Mean absolute percentage error values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.random(size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.mean_absolute_percentage_error(y_true, y_pred) rgY@rNr) rrrrepsilonrrrr)rrrdiffs rrBrBsL "6 * *F  "6 > > >F#GO$5$566G3FFCCNFF 7FVOs{376??G'L'LL M MD 38Dr*** **r)z,keras.metrics.mean_squared_logarithmic_errorz+keras.losses.mean_squared_logarithmic_errorzkeras._legacy.losses.mslezkeras._legacy.losses.MSLEzkeras._legacy.metrics.mslezkeras._legacy.metrics.MSLEctjtj}tj|}tj||j}t ||\}}tjtj||dz}tjtj||dz}tjtj ||z dS)a4Computes the mean squared logarithmic error between `y_true` & `y_pred`. Formula: ```python loss = mean(square(log(y_true + 1) - log(y_pred + 1)), axis=-1) ``` Note that `y_pred` and `y_true` cannot be less or equal to 0. Negative values and 0 values will be replaced with `keras.backend.epsilon()` (default to `1e-7`). Args: y_true: Ground truth values with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Mean squared logarithmic error values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.randint(0, 2, size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.mean_squared_logarithmic_error(y_true, y_pred) rrUrNr) rrrrrrlogrrr)rrr first_log second_logs rrIrIsL#GO$5$566G  "6 * *F  "6 > > >F3FFCCNFF FG44s:;;IVW55;<>> y_true = [[0., 1.], [1., 1.], [1., 1.]] >>> y_pred = [[1., 0.], [1., 1.], [-1., -1.]] >>> loss = keras.losses.cosine_similarity(y_true, y_pred, axis=-1) [-0., -0.99999994, 0.99999994] rr)rrrrr r)rrrQs rrOrO s@ "6 * *F  "6 > > >F3FFCCNFF vD ) ) )F vD ) ) )F GFVO$ / / / //rzkeras.losses.huberzkeras.metrics.huberrUc tj|}tj||j}t||\}}tj|}tj||}tj|}tjd|j}tjtj||k|tj|z||z|tj|zz dS)a Computes Huber loss value. Formula: ```python for x in error: if abs(x) <= delta: loss.append(0.5 * x^2) elif abs(x) > delta: loss.append(delta * abs(x) - 0.5 * delta^2) loss = mean(loss, axis=-1) ``` See: [Huber loss](https://en.wikipedia.org/wiki/Huber_loss). Example: >>> y_true = [[0, 1], [0, 0]] >>> y_pred = [[0.6, 0.4], [0.4, 0.6]] >>> loss = keras.losses.huber(y_true, y_pred) 0.155 Args: y_true: tensor of true targets. y_pred: tensor of predicted targets. delta: A float, the point where the Huber loss function changes from a quadratic to linear. Defaults to `1.0`. Returns: Tensor with one scalar loss entry per sample. r?rNr) rrrrsubtractrrwherer)rrrXerror abs_errorhalfs rrYrY3sB "6 * *F  "6 > > >F3FFCCNFF  !% ( (E L ( (EI  IO < < >> y_true = [[0., 1.], [0., 0.]] >>> y_pred = [[1., 1.], [0., 0.]] >>> loss = keras.losses.log_cosh(y_true, y_pred) 0.108 Args: y_true: Ground truth values with shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values with shape = `[batch_size, d0, .. dN]`. Returns: Logcosh error values with shape = `[batch_size, d0, .. dN-1]`. rrc>|tj|dzzz S)Ng)rsoftplus)xlog2s r_logcoshzlog_cosh.._logcoshs!3<D)))D00rrNr)rrrrrr)rrrrs @rr]r]esH "6 * *F  "6 > > >F3FFCCNFF  V\ B B BD11111 8HHVf_--B 7 7 77r)zkeras.metrics.kl_divergencezkeras.losses.kl_divergencezkeras._legacy.losses.KLDzkeras._legacy.losses.kldz0keras._legacy.losses.kullback_leibler_divergencezkeras._legacy.metrics.KLDzkeras._legacy.metrics.kldz1keras._legacy.metrics.kullback_leibler_divergencecVtj|}tj||j}tj|t jd}tj|t jd}tj|tj||z zdS)aComputes Kullback-Leibler divergence loss between `y_true` & `y_pred`. Formula: ```python loss = y_true * log(y_true / y_pred) ``` `y_true` and `y_pred` are expected to be probability distributions, with values between 0 and 1. They will get clipped to the `[0, 1]` range. Args: y_true: Tensor of true targets. y_pred: Tensor of predicted targets. Returns: KL Divergence loss values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.randint(0, 2, size=(2, 3)).astype(np.float32) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.kl_divergence(y_true, y_pred) >>> assert loss.shape == (2,) >>> y_true = ops.clip(y_true, 1e-7, 1) >>> y_pred = ops.clip(y_pred, 1e-7, 1) >>> assert np.array_equal( ... loss, np.sum(y_true * np.log(y_true / y_pred), axis=-1)) rrNr)rrrcliprrrrrs rrtrtsX "6 * *F  "66< 8 8F Xfgo// 3 3F Xfgo// 3 3F 76CGFVO4442 > > >>rzkeras.metrics.poissonzkeras.losses.poissonc tj|}tj||j}tjtj}tj||tj||zzz dS)aComputes the Poisson loss between y_true and y_pred. Formula: ```python loss = y_pred - y_true * log(y_pred) ``` Args: y_true: Ground truth values. shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values. shape = `[batch_size, d0, .. dN]`. Returns: Poisson loss values with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = np.random.randint(0, 2, size=(2, 3)) >>> y_pred = np.random.random(size=(2, 3)) >>> loss = keras.losses.poisson(y_true, y_pred) >>> assert loss.shape == (2,) >>> y_pred = y_pred + 1e-7 >>> assert np.allclose( ... loss, np.mean(y_pred - y_true * np.log(y_pred), axis=-1), ... atol=1e-5) rrNr)rrrrrrr)rrrs rryrysrB "6 * *F  "6 > > >F#GO$5$566G 8FVcgfw.>&?&???b I I IIrz&keras.metrics.categorical_crossentropyz%keras.losses.categorical_crossentropyFr~ct|tr"td|dt|t j|}t j||j}|jddkr%tj d|jdtd|r@t jt j|d|j}|d |z z||z z}t j |||| S) aComputes the categorical crossentropy loss. Args: y_true: Tensor of one-hot true targets. y_pred: Tensor of predicted targets. from_logits: Whether `y_pred` is expected to be a logits tensor. By default, we assume that `y_pred` encodes a probability distribution. label_smoothing: Float in [0, 1]. If > `0` then smooth the labels. For example, if `0.1`, use `0.1 / num_classes` for non-target labels and `0.9 + 0.1 / num_classes` for target labels. axis: Defaults to `-1`. The dimension along which the entropy is computed. Returns: Categorical crossentropy loss value. Example: >>> y_true = [[0, 1, 0], [0, 0, 1]] >>> y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]] >>> loss = keras.losses.categorical_crossentropy(y_true, y_pred) >>> assert loss.shape == (2,) >>> loss array([0.0513, 2.303], dtype=float32) -`axis` must be of type `int`. Received: axis= of type rNrzIn loss categorical_crossentropy, expected y_pred.shape to be (batch_size, num_classes) with num_classes > 1. Received: y_pred.shape=B. Consider using 'binary_crossentropy' if you only have 2 classes. stacklevelrUrrQ) isinstancebool ValueErrortyperrrrshapewarningswarn SyntaxWarningr)rrrrrQ num_classess rrrs0D$  :" : :-1$ZZ : :    "6 * *F Xffl + +F |B1  O `0` then smooth the labels. For example, if `0.1`, use `0.1 / num_classes` for non-target labels and `0.9 + 0.1 / num_classes` for target labels. axis: Defaults to `-1`. The dimension along which the entropy is computed. Returns: Categorical focal crossentropy loss value. Example: >>> y_true = [[0, 1, 0], [0, 0, 1]] >>> y_pred = [[0.05, 0.9, 0.05], [0.1, 0.85, 0.05]] >>> loss = keras.losses.categorical_focal_crossentropy(y_true, y_pred) >>> assert loss.shape == (2,) >>> loss array([2.63401289e-04, 6.75912094e-01], dtype=float32) rrrNrzIn loss categorical_focal_crossentropy, expected y_pred.shape to be (batch_size, num_classes) with num_classes > 1. Received: y_pred.shape=rrrrUrT)rQkeepdims)rrrrrrrrrrrrsoftmaxrrrrrpowermultiply) rrrrrrrQroutputccemodulating_factorweighting_factor focal_cces rrr-sb$  :" : :-1$ZZ : :    "6 * *F Xffl + +F |B1  O>> y_true = [1, 2] >>> y_pred = [[0.05, 0.95, 0], [0.1, 0.8, 0.1]] >>> loss = keras.losses.sparse_categorical_crossentropy(y_true, y_pred) >>> assert loss.shape == (2,) >>> loss array([0.0513, 2.303], dtype=float32) NrNrr~) rr not_equalrr expand_dimsrreshaper _keras_maskAttributeError)rrrrrQ res_shape valid_maskress rrrs JIf%%crc* ]638L&,+O+OPP #(:v|<<<#( OJ + +V\     -     C[Y77 i C-- (COO    D  JsC'' C43C4z!keras.metrics.binary_crossentropyz keras.losses.binary_crossentropyctj|}tj||j}|r|d|z zd|zz}tjtj||||S)aComputes the binary crossentropy loss. Args: y_true: Ground truth values. shape = `[batch_size, d0, .. dN]`. y_pred: The predicted values. shape = `[batch_size, d0, .. dN]`. from_logits: Whether `y_pred` is expected to be a logits tensor. By default, we assume that `y_pred` encodes a probability distribution. label_smoothing: Float in `[0, 1]`. If > `0` then smooth the labels by squeezing them towards 0.5, that is, using `1. - 0.5 * label_smoothing` for the target class and `0.5 * label_smoothing` for the non-target class. axis: The axis along which the mean is computed. Defaults to `-1`. Returns: Binary crossentropy loss value. shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = [[0, 1], [0, 0]] >>> y_pred = [[0.6, 0.4], [0.4, 0.6]] >>> loss = keras.losses.binary_crossentropy(y_true, y_pred) >>> assert loss.shape == (2,) >>> loss array([0.916 , 0.714], dtype=float32) rUr)rr)rrrrrr)rrrrrQs rrrs{D "6 * *F Xffl + +FJ301C/4II 8 KHHH    rz'keras.metrics.binary_focal_crossentropyz&keras.losses.binary_focal_crossentropyctj|}tj||j}|r|d|z zd|zz}|rtj|}tj||d}||zd|z d|z zz} tjd| z |} | |z} |r||zd|z d|z zz} | | z} tj| |S)aComputes the binary focal crossentropy loss. According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it helps to apply a focal factor to down-weight easy examples and focus more on hard examples. By default, the focal tensor is computed as follows: `focal_factor = (1 - output) ** gamma` for class 1 `focal_factor = output ** gamma` for class 0 where `gamma` is a focusing parameter. When `gamma` = 0, there is no focal effect on the binary crossentropy loss. If `apply_class_balancing == True`, this function also takes into account a weight balancing factor for the binary classes 0 and 1 as follows: `weight = alpha` for class 1 (`target == 1`) `weight = 1 - alpha` for class 0 where `alpha` is a float in the range of `[0, 1]`. Args: y_true: Ground truth values, of shape `(batch_size, d0, .. dN)`. y_pred: The predicted values, of shape `(batch_size, d0, .. dN)`. apply_class_balancing: A bool, whether to apply weight balancing on the binary classes 0 and 1. alpha: A weight balancing factor for class 1, default is `0.25` as mentioned in the reference. The weight for class 0 is `1.0 - alpha`. gamma: A focusing parameter, default is `2.0` as mentioned in the reference. from_logits: Whether `y_pred` is expected to be a logits tensor. By default, we assume that `y_pred` encodes a probability distribution. label_smoothing: Float in `[0, 1]`. If > `0` then smooth the labels by squeezing them towards 0.5, that is, using `1. - 0.5 * label_smoothing` for the target class and `0.5 * label_smoothing` for the non-target class. axis: The axis along which the mean is computed. Defaults to `-1`. Returns: Binary focal crossentropy loss value with shape = `[batch_size, d0, .. dN-1]`. Example: >>> y_true = [[0, 1], [0, 0]] >>> y_pred = [[0.6, 0.4], [0.4, 0.6]] >>> loss = keras.losses.binary_focal_crossentropy( ... y_true, y_pred, gamma=2) >>> assert loss.shape == (2,) >>> loss array([0.330, 0.206], dtype=float32) rUrF)targetrrrr)rrrrsigmoidrrr) rrrrrrrrQbcep_t focal_factor focal_bceweights rrrsB "6 * *F Xffl + +FJ301C/4II%V$$ !   C 6/QZAJ7 7C9S3Y..Ls"I'%1v:!e)"<<Y& 8ID ) ) ))rzkeras.losses.CTCc.eZdZdZ dfd ZdZxZS)CTCaCTC (Connectionist Temporal Classification) loss. Args: y_true: A tensor of shape `(batch_size, target_max_length)` containing the true labels in integer format. `0` always represents the blank/mask index and should not be used for classes. y_pred: A tensor of shape `(batch_size, output_max_length, num_classes)` containing logits (the output of your model). They should *not* be normalized via softmax. r ctccZtt||dSr_)rrrr4s rrz CTC.__init__es9        rc |j|jdSr_r`r7s rr zCTC.get_configpI   r)r rr9r/s@rrrXs^  (               rrzkeras.losses.ctccttj|dkr$tdtj|ttj|dkr$tdtj|tjtj|dd}tjtj|dd}tjtj|dd}|tj|fdz}|tj|fdz}tj||||d S) aCTC (Connectionist Temporal Classification) loss. Args: y_true: A tensor of shape `(batch_size, max_length)` containing the true labels in integer format. `0` always represents the blank/mask index and should not be used for classes. y_pred: A tensor of shape `(batch_size, max_length, num_classes)` containing logits (the output of your model). They should *not* be normalized via softmax. rz{Targets `y_true` are expected to be a tensor of shape `(batch_size, max_length)` in integer format. Received: y_true.shape=zuLogits `y_pred` are expected to be a tensor of shape `(batch_size, max_length, num_classes)`. Received: y_pred.shape=rint32rr) mask_index)lenrrrronesctc_loss)rr batch_length input_length label_lengths rrrwsa 39V  "" :&)i&7&7 : :    39V  "" :&)i&7&7 : :   8CIf--a0@@@L8CIf--a0@@@L8CIf--a0@@@L#(L?'"J"J"JJL#(L?'"J"J"JJL < lq   rzkeras.losses.Dicec.eZdZdZ dfd ZdZxZS)Dice3Computes the Dice loss value between `y_true` and `y_pred`. Formula: ```python loss = 1 - (2 * sum(y_true * y_pred)) / (sum(y_true) + sum(y_pred)) ``` Args: y_true: tensor of true targets. y_pred: tensor of predicted targets. Returns: Dice loss value. r dicecZtt||dSr_)rrr"r4s rrz Dice.__init__s9        rc |j|jdSr_r`r7s rr zDice.get_configrr)r r"r9r/s@rr r s^  "(               rr zkeras.losses.dicectj|}tj||j}tj|dg}tj|dg}tj||z}tjd|ztj|tj|ztjz}d|z S)r!rNrr rrrrrrdividerr)rrinputstargets intersectionr"s rr"r"s  "6 * *F Xffl + +F [" & &Fk&2$''G76G+,,L : l #'&//)GO,=,==  D t8Orzkeras.losses.Tverskyc2eZdZdZ dfd ZdZxZS)Tversky}Computes the Tversky loss value between `y_true` and `y_pred`. This loss function is weighted by the alpha and beta coefficients that penalize false positives and false negatives. With `alpha=0.5` and `beta=0.5`, the loss value becomes equivalent to Dice Loss. Args: y_true: tensor of true targets. y_pred: tensor of predicted targets. alpha: coefficient controlling incidence of false positives. beta: coefficient controlling incidence of false negatives. Returns: Tversky loss value. Reference: - [Salehi et al., 2017](https://arxiv.org/abs/1706.05721) rr tverskycztt||||||_||_dS)N)rbetarr)rrr.rr0)rrr0rrrs rrzTversky.__init__sI        rc8|j|j|j|jdS)Nrrr0rr2r7s rr zTversky.get_configs%IZI    r)rrr r.r9r/s@rr,r,sd0 '  "       rr,zkeras.losses.tverskyrctj|}tj||j}tj|dg}tj|dg}tj||z}tjd|z |z}tj|d|z z}tj||||zz||zztjz} d| z S)r-rNrr&) rrrr0r(r)r*fprr.s rr.r.s. "6 * *F Xffl + +F [" & &Fk&2$''G76G+,,L !g+' ( (B AJ' ( (BjrEz!BI-0A0AAG w;r)rN)rU)Fr~rN)rrFr~rN)FNrN)FrrFr~rN)rr)7r keras.srcrrkeras.src.api_exportrkeras.src.losses.lossrrkeras.src.savingrkeras.src.utils.numerical_utilsr r r1r<rArHrMrTr\rcrhrnrsrxr}rrrrrrdriror2r=rBrIrOrYr]rtryrrrrrrrr r"r,r.r&rrr:s ------&&&&&&@@@@@@......555555$2-..%%%%%*%%/.%2.//%%%%%+%%0/%2899%%%%%"5%%:9%:899%%%%%"5%%:9%:-..$%$%$%$%$%*$%$%/.$%N"##"%"%"%"%"% "%"%$#"%J$%%%%%%%!%%&%%2"##%%%%% %%$#%4)**%%%%%&%%+*%4-..%%%%%*%%/.%6)**%%%%%&%%+*%6$%%%%%%%!%%&%%./00l l l l l ,l l 10l ^455j j j j j 1j j 65j Z455V V V V V 1V V 65V r9::H H H H H #6H H ;:H V:;;O O O O O $7O O <;O d( FF  F>%$   @)( . .  .F  ::  :8  77  74   + +   +F  !A!A  !AH.//$0$0$00/$0N#%:;<<...=<.b#8#8#8L     #?#?  #?L JJ  JB0/BD666  6r65   WWW  Wt76@B777  7t+*BD%%%  %P10   T*T*T*  T*n !!       "! < !!  "! F!""       #" D!""#"<$%%. . . . . !. . &%. b$%%$$$&%$$$r