o ?en&@s.dZddlmZddlmZddlmZddlmZddlmZddlm Z ddlm Z dd lm Z dd lm Z dd l mZdd lmZd'ddZddZejddZedgdejjfddZedgdd ejjfddZedgdd(ddZedgdd)d d!Zed"gd # $ d*d%d&Zd S)+z0Utilities for manipulating the loss collections.)context) constant_op)dtypes)ops) array_ops) check_opscond)confusion_matrix)math_ops) tf_contextlib) tf_exportNc sj}|j}dur_j}|j}|dur-|dur-||dks$|ddkr,t\n2ttfddtdtdfdd}ttd|\durgfSj}|j} | dkrvfS|dur| dur| |dkrt dgn || dkrt dgfSt} | tfddfd d fd d } tt| dfd d| fS)aSqueeze or expand last dimension if needed. 1. Squeezes last dim of `y_pred` or `y_true` if their rank differs by 1 (using `confusion_matrix.remove_squeezable_dimensions`). 2. Squeezes or expands last dim of `sample_weight` if its rank differs by 1 from the new rank of `y_pred`. If `sample_weight` is scalar, it is kept scalar. This will use static shape if available. Otherwise, it will add graph operations, which could result in a performance hit. Args: y_pred: Predicted values, a `Tensor` of arbitrary dimensions. y_true: Optional label `Tensor` whose dimensions match `y_pred`. sample_weight: Optional weight scalar or `Tensor` whose dimensions match `y_pred`. Returns: Tuple of `y_pred`, `y_true` and `sample_weight`. Each of them possibly has the last dimension squeezed, `sample_weight` could be extended by one dimension. If `sample_weight` is None, (y_pred, y_true) is returned. Ncs tSN)r remove_squeezable_dimensionsy_predy_truerb/home/www/facesmatcher.com/pyenv/lib/python3.10/site-packages/tensorflow/python/ops/losses/util.pyHsz.squeeze_or_expand_dimensions..cstfddS)NcsfSrrrrrrrLsz@squeeze_or_expand_dimensions....rr) is_last_dim_1 squeeze_dimsrrrrrKsrctdgSNr)rsqueezer sample_weightrrrccs*fdd}ttd|fddS)Ncrr)r expand_dimsrrrrrfrzMsqueeze_or_expand_dimensions.._maybe_expand_weights..rcSrrrrrrrhr r equal)Zexpand_weights) rank_diffrrr_maybe_expand_weightses z;squeeze_or_expand_dimensions.._maybe_expand_weightscsttdS)Nrr#r)r&maybe_squeeze_weightsr%rr_maybe_adjust_weightsjs z;squeeze_or_expand_dimensions.._maybe_adjust_weightscr!rrrrrrrrr") shapeZndimsr rrrankr r$r rr ) rrrZ y_pred_shapeZ y_pred_rankZ y_true_shapeZ y_true_rankZmaybe_squeeze_dimsZ weights_shapeZ weights_rankZweights_rank_tensorr(r)r&rr'r%rrrrrsqueeze_or_expand_dimensionssP       r+cCs:t|tj}t|tj}t|d|\}}}t||S)aHScales loss values by the given sample weights. `sample_weight` dimensions are updated to match with the dimension of `losses` if possible by using squeeze/expand/broadcast. Args: losses: Loss tensor. sample_weight: Sample weights tensor. Returns: `losses` scaled by `sample_weight` with dtype float32. N)r castrZfloat32r+multiply)lossesr_rrrscale_losses_by_sample_weightws  r0ccs|jj}|dur|dkrtd|d|dVdSttjt|tj dt j dddg dVWddS1s@wYdS) zContext manager that checks that the rank of per_example_loss is at least 1. Args: per_example_loss: Per example loss tensor. Yields: A context manager. NrzoInvalid value passed for `per_example_loss`. Expected a tensor with at least rank 1. Received per_example_loss=z with rank r)ZdtypezTInvalid value passed for `per_example_loss`. Expected a tensor with at least rank 1.)message) r)r* ValueErrorrZcontrol_dependenciesrZassert_greater_equalrr r,rZint32)Zper_example_lossZ loss_rankrrrcheck_per_example_loss_ranks*  "r3zlosses.add_loss)v1cCs$|rtst||dSdSdS)zAdds a externally defined loss to the collection of losses. Args: loss: A loss `Tensor`. loss_collection: Optional collection to add the loss to. N)rZexecuting_eagerlyrZadd_to_collection)Zlossloss_collectionrrradd_losss r6zlosses.get_lossescCs t||S)zGets the list of losses from the loss_collection. Args: scope: An optional scope name for filtering the losses to return. loss_collection: Optional losses collection. Returns: a list of loss tensors. )rget_collection)scoper5rrr get_lossess r9z losses.get_regularization_lossescCsttjj|S)zGets the list of regularization losses. Args: scope: An optional scope name for filtering the losses to return. Returns: A list of regularization losses as Tensors. )rr7 GraphKeysZREGULARIZATION_LOSSESr8rrrget_regularization_lossess r<zlosses.get_regularization_losstotal_regularization_losscCs$t|}|r tj||dStdS)zGets the total regularization loss. Args: scope: An optional scope name for filtering the losses to return. name: The name of the returned tensor. Returns: A scalar regularization loss. nameg)r<r add_nrZconstant)r8r?r.rrrget_regularization_losss  rAzlosses.get_total_lossT total_losscCs*t|d}|r|t|d7}tj||dS)a~Returns a tensor whose value represents the total loss. In particular, this adds any losses you have added with `tf.add_loss()` to any regularization losses that have been added by regularization parameters on layers constructors e.g. `tf.layers`. Be very sure to use this if you are constructing a loss_op manually. Otherwise regularization arguments on `tf.layers` methods will not function. Args: add_regularization_losses: A boolean indicating whether or not to use the regularization losses in the sum. name: The name of the returned tensor. scope: An optional scope name for filtering the losses to return. Note that this filters the losses added with `tf.add_loss()` as well as the regularization losses to that scope. Returns: A `Tensor` whose value represents the total loss. Raises: ValueError: if `losses` is not iterable. r;r>)r9r<r r@)Zadd_regularization_lossesr?r8r.rrrget_total_losss rC)NNr)Nr=)TrBN)__doc__Ztensorflow.python.eagerrZtensorflow.python.frameworkrrrZtensorflow.python.opsrrr r r Ztensorflow.python.utilr Z tensorflow.python.util.tf_exportr r+r0contextmanagerr3r:ZLOSSESr6r9r<rArCrrrrs:            Y