Tensorflow.js tf.layers.simpleRNN() Function
The tf.layers.simpleRNN() function is used to create a RNN layer consisting of a single SimpleRNNCell.
Parameters: It accepts the args object which can have the following properties:
- units (number): The output space’s dimensions, expressed as a positive integer.
- activation: The layer’s activation function.
- useBias (boolean): If the layer has a bias vector or not. True is the default value.
- kernelInitializer: The convolutional kernel weights matrix’s initializer.
- recurrentInitializer: The recurrentKernel weights matrix’s initializer. It is used for the linear transformation of the recurrent state.
- biasInitializer: The bias vector’s initializer.
- kernelRegularizer: The regularizer function applied to the kernel weights matrix.
- recurrentRegularizer: The regularizer function applied to the recurrentKernel weights matrix.
- biasRegularizer: The regularizer function applied to the bias vector.
- kernelConstraint: The constraint for the convolutional kernel weights.
- recurrentConstraint: The constraint for the recurrentKernel weights.
- biasConstraint: The constraint for the bias vector.
- dropout (number): It is a number between 0 and 1. The fraction of the units to drop for the linear transformation of the inputs.
- recurrentDropout (number): It is a number between 0 and 1. The fraction of the units to drop for the linear transformation of the recurrent state.
- dropoutFunc: This is included for the purpose of testing DI.
- cell: A RNN cell instance.
- returnSequences (boolean): Whether the final output in the output series should be returned, or the entire sequence should be returned.
- returnState (boolean): Whether or not the last state should be returned along with the output.
- goBackwards (boolean): If this is true, process the input sequence backward and return the reversed sequence. The default value is false.
- stateful (boolean): If true, the final state of each sample at index I in a batch will be used as the beginning state of the next batch’s sample at index i (default: false).
- unroll (boolean): The network will be unrolled if true; else, a symbolic loop will be utilized. Although unrolling can speed up an RNN, it is more memory-intensive. Only short sequences are acceptable for unrolling (default: false).
- inputDim (number): The input’s dimensionality (integer). When this layer is used as the initial layer in a model, this option (or the option inputShape) is necessary.
- inputLength (number): When the length of the input sequences is constant, it must be given. If you want to link Flatten and Dense layers upstream, you’ll need this parameter (without it, the shape of the dense outputs cannot be computed).
- inputShape: If this property is set, it will be utilized to construct an input layer that will be inserted before this layer.
- batchInputShape: If this property is set, an input layer will be created and inserted before this layer.
- batchSize: If batchInputShape isn’t supplied and inputShape is, batchSize is utilized to build the batchInputShape.
- dtype: It is the kind of data type for this layer. float32 is the default value. This parameter applies exclusively to input layers.
- name: This is the layer’s name and is of string type.
- trainable: If the weights of this layer may be changed by fit. True is the default value.
- weights: The layer’s initial weight values.
- inputDType: It is used for Legacy support.
Returns: It returns an object (SimpleRNN).
Tensor [[[0.9078521, -0.9811671, 0.7162469, 0.9916067], [0.9999183, -0.9997805, 0.8239585, 0.9999147], [0.9999995, -0.9999998, 0.9744635, 0.9999991], [1 , -1 , 0.9965866, 1 ]]]
Tensor [[[0.2636383 , 0.9990318, 0.1660565, 0.9994429, -0.1762104, -0.9415753, 0.2943841, 0.7435381], [-0.9700606, 0.9999998, 0.5248303, 1 , -0.6762528, -0.9998503, 0.7585124, 0.836854 ], [-0.9959837, 1 , 0.5081902, 1 , -0.9194239, -0.9999997, 0.9733018, 0.988907 ], [-0.9993195, 1 , 0.8597047, 1 , -0.9791942, -1 , 0.9934399, 0.9968426], [-0.999855 , 1 , 0.9431108, 1 , -0.9907937, -1 , 0.9975212, 0.9990824]]]