o 2hl@sddlmZddlmZddlmZddlmZddlmZddlmZddlmZddlm Z dd l m Z dd l m Z dd lmZdd lmZe d Gddde eZe dGdddeZdS)) activations)backend) constraints) initializers)ops) regularizers)tree) keras_export) InputSpec)Layer)DropoutRNNCell)RNNzkeras.layers.LSTMCellcs~eZdZdZ              dfd d Zfd d ZddZddZdddZfddZ dddZ Z S)LSTMCellax Cell class for the LSTM layer. This class processes one step within the whole time sequence input, whereas `keras.layer.LSTM` processes the whole sequence. Args: units: Positive integer, dimensionality of the output space. activation: Activation function to use. Default: hyperbolic tangent (`tanh`). If you pass None, no activation is applied (ie. "linear" activation: `a(x) = x`). recurrent_activation: Activation function to use for the recurrent step. Default: sigmoid (`sigmoid`). If you pass `None`, no activation is applied (ie. "linear" activation: `a(x) = x`). use_bias: Boolean, (default `True`), whether the layer should use a bias vector. kernel_initializer: Initializer for the `kernel` weights matrix, used for the linear transformation of the inputs. Default: `"glorot_uniform"`. recurrent_initializer: Initializer for the `recurrent_kernel` weights matrix, used for the linear transformation of the recurrent state. Default: `"orthogonal"`. bias_initializer: Initializer for the bias vector. Default: `"zeros"`. unit_forget_bias: Boolean (default `True`). If `True`, add 1 to the bias of the forget gate at initialization. Setting it to `True` will also force `bias_initializer="zeros"`. This is recommended in [Jozefowicz et al.]( https://github.com/mlresearch/v37/blob/gh-pages/jozefowicz15.pdf) kernel_regularizer: Regularizer function applied to the `kernel` weights matrix. Default: `None`. recurrent_regularizer: Regularizer function applied to the `recurrent_kernel` weights matrix. Default: `None`. bias_regularizer: Regularizer function applied to the bias vector. Default: `None`. kernel_constraint: Constraint function applied to the `kernel` weights matrix. Default: `None`. recurrent_constraint: Constraint function applied to the `recurrent_kernel` weights matrix. Default: `None`. bias_constraint: Constraint function applied to the bias vector. Default: `None`. dropout: Float between 0 and 1. Fraction of the units to drop for the linear transformation of the inputs. Default: 0. recurrent_dropout: Float between 0 and 1. Fraction of the units to drop for the linear transformation of the recurrent state. Default: 0. seed: Random seed for dropout. Call arguments: inputs: A 2D tensor, with shape `(batch, features)`. states: A 2D tensor with shape `(batch, units)`, which is the state from the previous time step. training: Python boolean indicating whether the layer should behave in training mode or in inference mode. Only relevant when `dropout` or `recurrent_dropout` is used. Example: >>> inputs = np.random.random((32, 10, 8)) >>> rnn = keras.layers.RNN(keras.layers.LSTMCell(4)) >>> output = rnn(inputs) >>> output.shape (32, 4) >>> rnn = keras.layers.RNN( ... keras.layers.LSTMCell(4), ... return_sequences=True, ... return_state=True) >>> whole_sequence_output, final_state = rnn(inputs) >>> whole_sequence_output.shape (32, 10, 4) >>> final_state.shape (32, 4) tanhsigmoidTglorot_uniform orthogonalzerosNc sF|dkr td|d|dd}tjd i|||_||_t||_t||_ ||_ t ||_ t ||_ t ||_t| |_t| |_t| |_t| |_t| |_t||_tdtd||_tdtd||_|jdkrd|_|jdkrd |_||_tjj|d |_ ||_!|j|jg|_"|j|_#dS) NrzQReceived an invalid value for argument `units`, expected a positive integer, got .implementation?r)seed)$ ValueErrorpopsuper__init__runitsrget activationrecurrent_activationuse_biasrkernel_initializerrecurrent_initializerbias_initializerrkernel_regularizerrecurrent_regularizerbias_regularizerrkernel_constraintrecurrent_constraintbias_constraintminmaxdropoutrecurrent_dropoutdropout_mask_countrrrandom SeedGeneratorseed_generatorunit_forget_bias state_size output_size)selfr!r#r$r%r&r'r(r7r)r*r+r,r-r.r1r2rkwargsr __class__rT/var/www/html/chatgem/venv/lib/python3.10/site-packages/keras/src/layers/rnn/lstm.pyr XsB               zLSTMCell.__init__cst||d}j|jdfdjjjd_jjjdfdjj j d_ j rTj r>fdd}nj}jjdfd|jjd_dSd_dS) Nrkernel)shapename initializer regularizer constraintrecurrent_kernelcs`tjjfg|Ri|tdjfg|Ri|jjdfg|Ri|gS)Nonesr)r concatenater(r!rr")_argsr;r:rr>r(s0 z(LSTMCell.build..bias_initializerbias)rbuild add_weightr!r&r)r,r@r'r*r-rFr%r7r(r+r.rL)r: input_shape input_dimr(r<rKr>rMs8    zLSTMCell.buildc Cs|\}}}}|\}} } } ||t||jddd|jf} ||t| |jdd|j|jdf} | || ||t| |jdd|jd|jdf}||t| |jdd|jddf}||fS)z.Computes carry and output using split kernels.Nr)r$rmatmulrFr!r#)r:xh_tm1c_tm1x_ix_fx_cx_oh_tm1_ih_tm1_fh_tm1_ch_tm1_oifcorrr>_compute_carry_and_outputs4  "  "z"LSTMCell._compute_carry_and_outputc CsH|\}}}}||}||}|||||} ||} | | fS)z.Computes carry and output using fused kernels.)r$r#) r:zrUz0z1z2z3r^r_r`rarrr>_compute_carry_and_output_fuseds    z(LSTMCell._compute_carry_and_output_fusedFc!Cs@|d}|d}|jdkr|r:d|jkrdkr:nn||}||d}||d}||d} ||d} n|}|}|} |} tj|jddd\} } } }t|| }t|| }t| | }t| |}|jrtj|jddd\}}}}||7}||7}||7}||7}|rd|j krdkrnn| |}||d}||d}||d}||d}n|}|}|}|}||||f}||||f}| |||\}}nA|rd|jkrdkrnn ||}||}t||j}|t||j 7}|jr||j7}tj|ddd}| ||\}}|||} | | |gfS) NrrrrrrQraxis)rr1get_dropout_maskrsplitr@rRr%rLr2get_recurrent_dropout_maskrbrFrhr#)!r:inputsstatestrainingrTrUdp_maskinputs_iinputs_finputs_cinputs_ok_ik_fk_ck_orVrWrXrYb_ib_fb_cb_o rec_dp_maskrZr[r\r]rSr`rarchrrr>calls`                  z LSTMCell.callcsid|jdt|jdt|jd|jd|jdt|jdt|j dt|j d t |j d t |j d t |jd t|jd t|jdt|jd|jd|jd|j}t}i||S)Nr!r#r$r%r7r&r'r(r)r*r+r,r-r.r1r2r)r!r serializer#r$r%r7rr&r'r(rr)r*r+rr,r-r.r1r2rr get_configr:config base_configr<rr>r!sb        zLSTMCell.get_configcsfddjDS)Ncs g|] }tj|fjdqS))dtype)rr compute_dtype).0d batch_sizer:rr> Esz.LSTMCell.get_initial_state..)r8)r:rrrr>get_initial_stateDs zLSTMCell.get_initial_state)rrTrrrTNNNNNNrrNFN) __name__ __module__ __qualname____doc__r rMrbrhrrr __classcell__rrr<r>rs2J ;0 ;#rzkeras.layers.LSTMcsDeZdZdZ                d6fd d Zd7fdd Zd8fdd ZeddZeddZ eddZ eddZ eddZ eddZ eddZed d!Zed"d#Zed$d%Zed&d'Zed(d)Zed*d+Zed,d-Zed.d/Zed0d1Zfd2d3Zed4d5ZZS)9LSTMa@Long Short-Term Memory layer - Hochreiter 1997. Based on available runtime hardware and constraints, this layer will choose different implementations (cuDNN-based or backend-native) to maximize the performance. If a GPU is available and all the arguments to the layer meet the requirement of the cuDNN kernel (see below for details), the layer will use a fast cuDNN implementation when using the TensorFlow backend. The requirements to use the cuDNN implementation are: 1. `activation` == `tanh` 2. `recurrent_activation` == `sigmoid` 3. `recurrent_dropout` == 0 4. `unroll` is `False` 5. `use_bias` is `True` 6. Inputs, if use masking, are strictly right-padded. 7. Eager execution is enabled in the outermost context. For example: >>> inputs = np.random.random((32, 10, 8)) >>> lstm = keras.layers.LSTM(4) >>> output = lstm(inputs) >>> output.shape (32, 4) >>> lstm = keras.layers.LSTM( ... 4, return_sequences=True, return_state=True) >>> whole_seq_output, final_memory_state, final_carry_state = lstm(inputs) >>> whole_seq_output.shape (32, 10, 4) >>> final_memory_state.shape (32, 4) >>> final_carry_state.shape (32, 4) Args: units: Positive integer, dimensionality of the output space. activation: Activation function to use. Default: hyperbolic tangent (`tanh`). If you pass `None`, no activation is applied (ie. "linear" activation: `a(x) = x`). recurrent_activation: Activation function to use for the recurrent step. Default: sigmoid (`sigmoid`). If you pass `None`, no activation is applied (ie. "linear" activation: `a(x) = x`). use_bias: Boolean, (default `True`), whether the layer should use a bias vector. kernel_initializer: Initializer for the `kernel` weights matrix, used for the linear transformation of the inputs. Default: `"glorot_uniform"`. recurrent_initializer: Initializer for the `recurrent_kernel` weights matrix, used for the linear transformation of the recurrent state. Default: `"orthogonal"`. bias_initializer: Initializer for the bias vector. Default: `"zeros"`. unit_forget_bias: Boolean (default `True`). If `True`, add 1 to the bias of the forget gate at initialization. Setting it to `True` will also force `bias_initializer="zeros"`. This is recommended in [Jozefowicz et al.]( https://github.com/mlresearch/v37/blob/gh-pages/jozefowicz15.pdf) kernel_regularizer: Regularizer function applied to the `kernel` weights matrix. Default: `None`. recurrent_regularizer: Regularizer function applied to the `recurrent_kernel` weights matrix. Default: `None`. bias_regularizer: Regularizer function applied to the bias vector. Default: `None`. activity_regularizer: Regularizer function applied to the output of the layer (its "activation"). Default: `None`. kernel_constraint: Constraint function applied to the `kernel` weights matrix. Default: `None`. recurrent_constraint: Constraint function applied to the `recurrent_kernel` weights matrix. Default: `None`. bias_constraint: Constraint function applied to the bias vector. Default: `None`. dropout: Float between 0 and 1. Fraction of the units to drop for the linear transformation of the inputs. Default: 0. recurrent_dropout: Float between 0 and 1. Fraction of the units to drop for the linear transformation of the recurrent state. Default: 0. seed: Random seed for dropout. return_sequences: Boolean. Whether to return the last output in the output sequence, or the full sequence. Default: `False`. return_state: Boolean. Whether to return the last state in addition to the output. Default: `False`. go_backwards: Boolean (default: `False`). If `True`, process the input sequence backwards and return the reversed sequence. stateful: Boolean (default: `False`). If `True`, the last state for each sample at index i in a batch will be used as initial state for the sample of index i in the following batch. unroll: Boolean (default False). If `True`, the network will be unrolled, else a symbolic loop will be used. Unrolling can speed-up a RNN, although it tends to be more memory-intensive. Unrolling is only suitable for short sequences. use_cudnn: Whether to use a cuDNN-backed implementation. `"auto"` will attempt to use cuDNN when feasible, and will fallback to the default implementation if not. Call arguments: inputs: A 3D tensor, with shape `(batch, timesteps, feature)`. mask: Binary tensor of shape `(samples, timesteps)` indicating whether a given timestep should be masked (optional). An individual `True` entry indicates that the corresponding timestep should be utilized, while a `False` entry indicates that the corresponding timestep should be ignored. Defaults to `None`. training: Python boolean indicating whether the layer should behave in training mode or in inference mode. This argument is passed to the cell when calling it. This is only relevant if `dropout` or `recurrent_dropout` is used (optional). Defaults to `None`. initial_state: List of initial state tensors to be passed to the first call of the cell (optional, `None` causes creation of zero-filled initial state tensors). Defaults to `None`. rrTrrrNrFautoc s(t|fid|d|d|d|d|d|d|d| d | d | d | d |d |d|d|d|ddd|ddddd|d|dd}tj|f|||||| d|tdd|_|dvrptd|||_t dkrt |j |j |j |jr|dvrd|_dSdSdSdS) Nr#r$r%r&r7r'r(r)r*r+r,r-r.r1r2r trainableTrB lstm_cellrrr)return_sequences return_state go_backwardsstatefulunrollactivity_regularizerrQ)ndim)rTFzlInvalid valid received for argument `use_cudnn`. Expected one of {'auto', True, False}. Received: use_cudnn= tensorflow)TrF)rr"rrr r input_specr use_cudnnrcudnn_okr#r$rr% supports_jit)r:r!r#r$r%r&r'r(r7r)r*r+rr,r-r.r1r2rrrrrrrr;cellr<rr>r s           z LSTM.__init__cst|r |d}|jdvrt|jstzY|r;|jr;|j|dddddf}tj|dd}t |t |}||}n|}t j ||d|d||jj |jj|jj|jj|jj|j|j|jd }t dkrhd|_|WStysYnw|jdur}td tj||||d S) Nr)rTrri)r@rFrLr#r$rrrrFTzuse_cudnn=True was specified, but cuDNN is not supported for this layer configuration with this backend. Pass use_cudnn='auto' to fallback to a non-cuDNN implementation.)maskrp)r is_nestedrr2r1rrkr expand_dims broadcast_torArlstmr@rFrLr#r$rrrrNotImplementedErrorrr inner_loop)r: sequences initial_staterrprq dp_sequencesoutr<rr>rsP        zLSTM.inner_loopcstj||||dS)N)rrpr)rr)r:rrrrpr<rr>rFsz LSTM.callcC|jjSr)rr!rKrrr>r!Kz LSTM.unitscCrr)rr#rKrrr>r#OrzLSTM.activationcCrr)rr$rKrrr>r$SrzLSTM.recurrent_activationcCrr)rr%rKrrr>r%Wrz LSTM.use_biascCrr)rr7rKrrr>r7[rzLSTM.unit_forget_biascCrr)rr&rKrrr>r&_rzLSTM.kernel_initializercCrr)rr'rKrrr>r'crzLSTM.recurrent_initializercCrr)rr(rKrrr>r(grzLSTM.bias_initializercCrr)rr)rKrrr>r)krzLSTM.kernel_regularizercCrr)rr*rKrrr>r*orzLSTM.recurrent_regularizercCrr)rr+rKrrr>r+srzLSTM.bias_regularizercCrr)rr,rKrrr>r,wrzLSTM.kernel_constraintcCrr)rr-rKrrr>r-{rzLSTM.recurrent_constraintcCrr)rr.rKrrr>r.rzLSTM.bias_constraintcCrr)rr1rKrrr>r1rz LSTM.dropoutcCrr)rr2rKrrr>r2rzLSTM.recurrent_dropoutcsid|jdt|jdt|jd|jdt|jdt|jdt|j d|j d t |j d t |j d t |jd t |jd t|jdt|jdt|jd|jd|jd|jji}t}|d=i||S)Nr!r#r$r%r&r'r(r7r)r*r+rr,r-r.r1r2rr)r!rrr#r$r%rr&r'r(r7rr)r*r+rrr,r-r.r1r2rrrrrr<rr>rsl        " zLSTM.get_configcCs|di|S)Nrr)clsrrrr> from_configszLSTM.from_config)rrTrrrTNNNNNNNrrNFFFFFrr)NNF)rrrrr rrpropertyr!r#r$r%r7r&r'r(r)r*r+r,r-r.r1r2r classmethodrrrrr<r>rKs~vQ5                 'rN) keras.srcrrrrrrrkeras.src.api_exportr keras.src.layers.input_specr keras.src.layers.layerr %keras.src.layers.rnn.dropout_rnn_cellr keras.src.layers.rnn.rnnr rrrrrr>s$           =