o i eX@sdZddlZddlmmZddlmZddlm Z ddl m Z ddl m Z ddlmZddlmZdd lmZdd lmZdd lmZdd lmZed GdddeZdS)zBidirectional wrapper for RNNs.N)backend)Layer) InputSpec) rnn_utils)Wrapper)serialization_lib) generic_utils) tf_inspect)tf_utils) keras_exportzkeras.layers.BidirectionalcseZdZdZ   dfdd ZeddZdd Zd d d Ze j d dZ d!fdd Z    d"ddZ d#ddZddZddZeddZfddZed#ddZZS)$ BidirectionalaY Bidirectional wrapper for RNNs. Args: layer: `keras.layers.RNN` instance, such as `keras.layers.LSTM` or `keras.layers.GRU`. It could also be a `keras.layers.Layer` instance that meets the following criteria: 1. Be a sequence-processing layer (accepts 3D+ inputs). 2. Have a `go_backwards`, `return_sequences` and `return_state` attribute (with the same semantics as for the `RNN` class). 3. Have an `input_spec` attribute. 4. Implement serialization via `get_config()` and `from_config()`. Note that the recommended way to create new RNN layers is to write a custom RNN cell and use it with `keras.layers.RNN`, instead of subclassing `keras.layers.Layer` directly. - When the `returns_sequences` is true, the output of the masked timestep will be zero regardless of the layer's original `zero_output_for_mask` value. merge_mode: Mode by which outputs of the forward and backward RNNs will be combined. One of {'sum', 'mul', 'concat', 'ave', None}. If None, the outputs will not be combined, they will be returned as a list. Default value is 'concat'. backward_layer: Optional `keras.layers.RNN`, or `keras.layers.Layer` instance to be used to handle backwards input processing. If `backward_layer` is not provided, the layer instance passed as the `layer` argument will be used to generate the backward layer automatically. Note that the provided `backward_layer` layer should have properties matching those of the `layer` argument, in particular it should have the same values for `stateful`, `return_states`, `return_sequences`, etc. In addition, `backward_layer` and `layer` should have different `go_backwards` argument values. A `ValueError` will be raised if these requirements are not met. Call arguments: The call arguments for this layer are the same as those of the wrapped RNN layer. Beware that when passing the `initial_state` argument during the call of this layer, the first half in the list of elements in the `initial_state` list will be passed to the forward RNN call and the last half in the list of elements will be passed to the backward RNN call. Raises: ValueError: 1. If `layer` or `backward_layer` is not a `Layer` instance. 2. In case of invalid `merge_mode` argument. 3. If `backward_layer` has mismatched properties compared to `layer`. Examples: ```python model = Sequential() model.add(Bidirectional(LSTM(10, return_sequences=True), input_shape=(5, 10))) model.add(Bidirectional(LSTM(10))) model.add(Dense(5)) model.add(Activation('softmax')) model.compile(loss='categorical_crossentropy', optimizer='rmsprop') # With custom backward layer model = Sequential() forward_layer = LSTM(10, return_sequences=True) backward_layer = LSTM(10, activation='relu', return_sequences=True, go_backwards=True) model.add(Bidirectional(forward_layer, backward_layer=backward_layer, input_shape=(5, 10))) model.add(Dense(5)) model.add(Activation('softmax')) model.compile(loss='categorical_crossentropy', optimizer='rmsprop') ``` concatNc sft|ts td||durt|tstd||dvr(td|dd|_tj|fi|d|_|||_|durK|j|dd|_n ||_t ||_ d |jj |j_ d |jj |j_ |d d }||j||j||_|rt|}|d|d |j_||d d|j_|j|_|j|_|j|_d|_|d|j|_d|_|j|_dS)Nz[Please initialize `Bidirectional` layer with a `tf.keras.layers.Layer` instance. Received: zJ`backward_layer` need to be a `tf.keras.layers.Layer` instance. Received: )summulaver NzInvalid merge mode. Received: zC. Merge mode should be one of {"sum", "mul", "ave", "concat", None}FT) go_backwardsforward_ backward_cSs t|dddur|j|_dSdS)Nzero_output_for_mask)getattrreturn_sequencesr)layerrQ/var/www/myenv/lib/python3.10/site-packages/keras/src/layers/rnn/bidirectional.pyforce_zero_output_for_masks z:Bidirectional.__init__..force_zero_output_for_mask trainabler) isinstancer ValueError_setattr_trackingsuper__init___recreate_layer_from_config forward_layerbackward_layerrserialize_keras_object_backward_layer_configname_name_verify_layer_config merge_modeleninitial_weightsstatefulr return_statesupports_maskinggetr _trainable_num_constants input_spec)selfrr*weightsr$kwargsrnw __class__rrr!msZ       zBidirectional.__init__cCs|jjSN)r!_use_input_spec_as_call_signature)r4rrrr;sz/Bidirectional._use_input_spec_as_call_signaturec Csz|jj|jjkrtd|jjd|jjd}|D] }t|j|}t|j|}||kr:td|d|d|dqdS) zBEnsure the forward and backward layers have valid common property.ziForward layer and backward layer should have different `go_backwards` value.forward_layer.go_backwards = z,backward_layer.go_backwards = )r-rr.zTForward layer and backward layer are expected to have the same value for attribute "z", got "z" for forward layer and "z" for backward layerN)r#rr$rr)r4common_attributesa forward_valuebackward_valuerrrr)s0   z"Bidirectional._verify_layer_configFcCs|}|r |d |d<dt|jjjvrDi}t|dd}|dur<|j||jj<t|dg}|D] }|j||jj<q2|jj||dS|j|S)Nrcustom_objectscellcellsr@) get_configr getfullargspecr9 from_configargsr__name__)r4rrconfigr@rA stacked_cellscrrrr"s"   z)Bidirectional._recreate_layer_from_configcCs|j|}|jrtj|dddd}tj|ddd}ntj|dd}|jdkr<|}|dd9<t|}n |jdurH|t |g}|jrc|jdurY||t |S|g|t |S|S)NF) to_tuplesrr r) r#compute_output_shaper.r convert_shapesr*as_listtf TensorShapecopy)r4 input_shape output_shape state_shaperrrrOs,      z"Bidirectional.compute_output_shapec st||||j\}}}t|tr!t|dkr|dd}|d}|dur3|dur3tj|fi|Sg}g}|durut|}|ddkrLtd|||d<||7}t j dd|}|d|d|j _ ||dd|j_ ||7}|dur||d <||7}d d |D} | |j _| |j_|| 7}t||_|j|j _|j|j_tt j |d} t j |D] } t| | krtd q| r|g|} d d ttt j |D|} d|d<d|d <|j}| |_tj| fi|}||_|Stj|fi|S)zN`Bidirectional.__call__` implements the same API as the wrapped `RNN`.rLNrrzWhen passing `initial_state` to a Bidirectional RNN, the state should be a list containing the states of the underlying RNNs. Received: initial_statecSstt|dS)Nshaperr int_shape)staterrr*z(Bidirectional.__call__.. constantscSsg|] }tt|dqS)rYr[).0constantrrr 3sz*Bidirectional.__call__..zThe initial state of a Bidirectional layer cannot be specified with a mix of Keras tensors and non-Keras tensors (a "Keras tensor" is a tensor that was returned by a Keras layer, or by `Input`)cSg|]}dqSr:rra_rrrrcQs)rstandardize_argsr2rlistr+r __call__rrRnest map_structurer# state_specr$constants_specris_keras_tensorflattenranger3)r4inputsrXr`r6additional_inputsadditional_specs num_states state_specsrmrntensor full_inputfull_input_specoriginal_input_specoutputr8rrri s         zBidirectional.__call__cCsi}t|jjdr||d<t|jjdr||d<t|jjdr&||d<t|jjdrt|trt|dkr|dg}|dg}t||jdd} ||d| 7}|jsb||| d7}n||| |j 7}|||j d7}|||j d7}d \} } d|vrd|d<n%|dur||}}t|d} |d| } || d} n ||}}d \} } |j|fd| i|} |j |fd| i|}n|j|fi|} |j |fi|}|j r| dd|dd}| d} |d}|j rt |jd d rdnd}t ||}|jd krt | |g}n7|jd kr | |}n,|jdkr-| |d}n|jdkr8| |}n|jdurC| |g}n td|jd|j r_|jdurZ||S|g|S|S)zB`Bidirectional.call` implements the same API as the wrapped `RNN`.trainingmaskr`rXrLrrNNN time_majorFr rrrz/Unrecognized value for `merge_mode`. Received: z3Expected values are ["concat", "sum", "ave", "mul"])rhas_argrcallrrhr+r2r#r$r.rrrreverser* concatenater)r4rqr{r|rXr`r6forward_inputsbackward_inputspivot forward_statebackward_statehalfyy_revstatestime_dimrzrrrras                  zBidirectional.callcCs|jstd|dur|j|jdSt|ttfs%td|t |d}|j|d||j||ddS)NzLayer must be stateful.zRUnrecognized value for `states`. Expected `states` to be list or tuple. Received: r) r-AttributeErrorr# reset_statesr$rrhtuplerr+)r4rrrrrrs  zBidirectional.reset_statescCs~t|jj|j|Wdn1swYt|jj|j|Wdn1s5wYd|_dS)NT)r name_scoper#r'buildr$built)r4rUrrrrs zBidirectional.buildcCst|tr |d}|jr|js||g}n |}n |jsddgnd}|jr@|jj}dd|D}t|tr9||dS|g|dS|S)NrcSrdr:rrerrrrcr_z.Bidirectional.compute_mask..r)rrhrr*r.r#r)r4rqr| output_maskr state_maskrrr compute_masks    zBidirectional.compute_maskcCs0i}t|jdr||jj||jj|S)N constraints)hasattrr#updaterr$)r4rrrrrs  zBidirectional.constraintscsTd|ji}|jr |j|d<t|dr|j|d<t}tt|t|S)Nr* num_constantsr&r$) r*r2rr&r rDdictrhitems)r4rI base_configr8rrrDs     zBidirectional.get_configcCsvt|}|dd}ddlm}||d|d|d<|dd}|dur/|||d}||d<|di|}||_|S)Nrr) deserializerrCr$r)rTdeepcopypopkeras.src.layersrr2)clsrIr@rdeserialize_layerbackward_layer_configr$rrrrrFs    zBidirectional.from_config)r NN)Fr})NNNNr:)rH __module__ __qualname____doc__r!propertyr;r)r"r shape_type_conversionrOrirrrrrrD classmethodrF __classcell__rrr8rr $s4JJ   Z `   r )rrTtensorflow.compat.v2compatv2rR keras.srcrkeras.src.engine.base_layerrkeras.src.engine.input_specrkeras.src.layers.rnnr!keras.src.layers.rnn.base_wrapperrkeras.src.savingrkeras.src.utilsrr r tensorflow.python.util.tf_exportr r rrrrs