tensorflow js加载gru模型

Question

我有一个基于 GRU 的模型。我已将其转换为 tensorflow js 。在 tfjs 中加载它时出现错误

未处理的拒绝（错误）：GRUCell 不支持将 reset_after 参数设置为 true。

我已经附上了相应的 json 。请指教

{"format": "layers-model", "generatedBy": "keras v2.4.0", "convertedBy": "TensorFlow.js Converter v2.7.0", "modelTopology": {"keras_version": "2.4 .0”，“后端”：“tensorflow”，“model_config”：{“class_name”：“Sequential”，“config”：{“name”：“sequential”，“layers”：[{“class_name”：“InputLayer ", "config": {"batch_input_shape": [null, 48, 64], "dtype": "float32", "sparse": false, "ragged": false, "name": "input_1"}}, { “class_name”：“LayerNormalization”，“config”：{“name”：“layer_normalization”，“trainable”：true，“dtype”：“float32”，“axis”：[2]，“epsilon”：0.001，“中心”：真，“规模”：真，“beta_initializer”：{“class_name”：“Zeros”，“config”：{}}，“gamma_initializer”：{“class_name”：“Ones”，“config”：{ }}，“beta_regularizer”：null，“gamma_regularizer”：null，“beta_constraint”：null，“gamma_constraint”：null}}，{“class_name”：“Conv1D”，“config”：{“name”：“conv1d” ，“可训练”：true，“dtype”：“float32”，“filters”：32，“kernel_size”：[3]，“strides”：[1]，“padding”：“valid”，“data_format “：“channels_last”，“dilation_rate”：[1]，“groups”：1，“activation”：“relu”，“use_bias”：true，“kernel_initializer”：{“class_name”：“GlorotUniform”，“config” ：{“seed”：null}}，“bias_initializer”：{“class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”：null，“bias_regularizer”：null，“activity_regularizer”：null，“ kernel_constraint”：null，“bias_constraint”：null}}，{“class_name”：“BatchNormalization”，“config”：{“name”：“batch_normalization”，“trainable”：true，“dtype”：“float32”，“轴”：[2]，“动量”：0.99，“epsilon”：0.001，“中心”：真，“规模”：真，“beta_initializer”：{“class_name”：“Zeros”，“config”：{} }，“gamma_initializer”：{“class_name”：“Ones”，“config”：{}}，“moving_mean_initializer”：{“class_name”：“Zeros”，“config”：{}}，“moving_variance_initializer”：{“ class_name": "Ones", "config": {}}, "beta_regularizer": null, "gamma_regularizer": null, "beta_constraint": null, "gamma_constraint": null}}, {"class_name": "MaxPooling1D", “配置”：{“名称”：“max_pooling1d”，“tr ainable”：true，“dtype”：“float32”，“strides”：[3]，“pool_size”：[3]，“padding”：“valid”，“data_format”：“channels_last”}}，{“class_name “：“Conv1D”，“config”：{“name”：“conv1d_1”，“trainable”：true，“dtype”：“float32”，“filters”：32，“kernel_size”：[3]，“strides” ：[1]，“填充”：“有效”，“data_format”：“channels_last”，“dilation_rate”：[1]，“组”：1，“激活”：“relu”，“use_bias”：true，“ kernel_initializer”：{“class_name”：“GlorotUniform”，“config”：{“seed”：null}}，“bias_initializer”：{“class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”： null，“bias_regularizer”：null，“activity_regularizer”：null，“kernel_constraint”：null，“bias_constraint”：null}}，{“class_name”：“Dropout”，“config”：{“name”：“dropout”， “可训练”：true，“dtype”：“float32”，“rate”：0.25，“noise_shape”：null，“seed”：null}}，{“class_name”：“GRU”，“config”：{“name “：“gru”，“可训练”：真，“dtype”：“float32”，“return_sequences”：真，“return_state”：假，“go_backwards”：假，“有状态”：假，“ unroll": false, "time_major": false, "units": 40, "activation": "tanh", "recurrent_activation": "sigmoid", "use_bias": true, "kernel_initializer": {"class_name": "GlorotUniform ", "config": {"seed": null}}, "recurrent_initializer": {"class_name": "Orthogonal", "config": {"gain": 1.0, "seed": null}}, "bias_initializer" ：{“class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”：null，“recurrent_regularizer”：null，“bias_regularizer”：null，“activity_regularizer”：null，“kernel_constraint”：null，“recurrent_constraint ": null, "bias_constraint": null, "dropout": 0.0, "recurrent_dropout": 0.0, "implementation": 2, "reset_after": true}}, {"class_name": "GRU", "config": { “名称”：“gru_1”，“可训练”：真，“dtype”：“float32”，“return_sequences”：假，“return_state”：假，“go_backwards”：假，“有状态”：假，“展开”： false，“time_major”：false，“units”：40，“activation”：“tanh”，“recurrent_activation”：“sigmoid”，“use_bias”：true，“kernel_initializer”：{“class_name”：“GlorotUniform ", "config": {"seed": null}}, "recurrent_initializer": {"class_name": "Orthogonal", "config": {"gain": 1.0, "seed": null}}, "bias_initializer" ：{“class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”：null，“recurrent_regularizer”：null，“bias_regularizer”：null，“activity_regularizer”：null，“kernel_constraint”：null，“recurrent_constraint ": null, "bias_constraint": null, "dropout": 0.0, "recurrent_dropout": 0.0, "implementation": 2, "reset_after": true}}, {"class_name": "Flatten", "config": { “名称”：“扁平化”，“可训练”：真，“dtype”：“float32”，“data_format”：“channels_last”}}，{“class_name”：“密集”，“config”：{“name”： “密集”，“可训练”：真，“dtype”：“float32”，“单位”：100，“激活”：“relu”，“use_bias”：真，“kernel_initializer”：{“class_name”：“GlorotUniform” ，“配置”：{“种子”：null}}，“bias_initializer”：{“class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”：null，“bias_regularizer”：null，“activity_regularizer” ：空，“kernel_constraint”：空，“bias_con应变“：null}}，{“class_name”：“Dropout”，“config”：{“name”：“dropout_1”，“trainable”：true，“dtype”：“float32”，“rate”：0.25，“噪声形状”：空，“种子”：空}}，{“类名”：“密集”，“配置”：{“名称”：“密集_1”，“可训练”：真，“dtype”：“float32”，“单位”：2，“激活”：“线性”，“use_bias”：true，“kernel_initializer”：{“class_name”：“GlorotUniform”，“config”：{“seed”：null}}，“bias_initializer”：{ “class_name”：“Zeros”，“config”：{}}，“kernel_regularizer”：null，“bias_regularizer”：null，“activity_regularizer”：null，“kernel_constraint”：null，“bias_constraint”：null}}]}} , "training_config": {"loss": {"class_name": "SparseCategoricalCrossentropy", "config": {"reduction": "auto", "name": "sparse_categorical_crossentropy", "from_logits": true}}, "metrics “：[“准确度”]，“weighted_metrics”：null，“loss_weights”：null，“optimizer_config”：{“class_name”：“Adam”，“config”：{“name”：“Adam”，“learning_rate”： 0.0010000000474974513，“衰减”：0.0，“beta_1”：0.8999999761581421，“beta _2": 0.9990000128746033, "epsilon": 1e-07, "amsgrad": false}}}}, "weightsManifest": [{"paths": ["group1-shard1of1.bin"], "weights": [{"名称”：“batch_normalization/gamma”，“shape”：[32]，“dtype”：“float32”}，{“name”：“batch_normalization/beta”，“shape”：[32]，“dtype”：“ float32"}，{"name"："batch_normalization/moving_mean"，"shape"：[32]，"dtype"："float32"}，{"name"："batch_normalization/moving_variance"，"shape"：[32] , "dtype": "float32"}, {"name": "conv1d/kernel", "shape": [3, 64, 32], "dtype": "float32"}, {"name": "conv1d/偏差”，“形状”：[32]，“dtype”：“float32”}，{“名称”：“conv1d_1/kernel”，“形状”：[3,32,32]，“dtype”：“float32” }, {"name": "conv1d_1/bias", "shape": [32], "dtype": "float32"}, {"name": "dense/kernel", "shape": [40, 100] ，“dtype”：“float32”}，{“name”：“dense/bias”，“shape”：[100]，“dtype”：“float32”}，{“name”：“dense_1/kernel”，“ shape”：[100, 2]，“dtype”：“float32”}，{“name”：“dense_1/bias”，“shape”：[2]，“dtype”：“float32”}，{“name” ：“gru/gru_cell/内核”，“形状”： [32, 120], "dtype": "float32"}, {"name": "gru/gru_cell/recurrent_kernel", "shape": [40, 120], "dtype": "float32"}, {"name ": "gru/gru_cell/bias", "shape": [2, 120], "dtype": "float32"}, {"name": "gru_1/gru_cell_1/kernel", "shape": [40, 120 ]，“dtype”：“float32”}，{“name”：“gru_1/gru_cell_1/recurrent_kernel”，“shape”：[40, 120]，“dtype”：“float32”}，{“name”：“gru_1 /gru_cell_1/bias", "shape": [2, 120], "dtype": "float32"}, {"name": "layer_normalization/gamma", "shape": [64], "dtype": "float32 "}, {"name": "layer_normalization/beta", "shape": [64], "dtype": "float32"}]}]}

Answer 1

我有同样的问题。在 python 上训练模型时，我尝试完全按照异常建议的方式进行操作。 修改了单元格初始化：

memory_layer = tf.keras.layers.GRUCell(units=memory_size, name='memory')

到

memory_layer = tf.keras.layers.GRUCell(units=memory_size, reset_after=False, name='memory')

在使用更改重新训练模型并导出到 tfjs 后，我能够加载它。此标志更改了单元内部架构，但在我的情况下，它不会对模型的性能产生影响。