javascript - 未知的正则化器 : L2 in tensorflowjs

Question

我已经使用模型在 python 中训练了一个模型

reg = 0.000001
model = Sequential()
model.add(Dense(24, activation='tanh', name='input_dense', input_shape=input_shape))
model.add(GRU(24, activation='tanh', recurrent_activation='sigmoid', return_sequences=True, kernel_regularizer=regularizers.l2(reg), recurrent_regularizer=regularizers.l2(reg), reset_after=False))
model.add(Flatten())
model.add(Dense(2, activation='softmax'))

但是当我使用“tensorflowjs_converter --input_format keras”转换这个模型并在浏览器中加载时出现错误

Unhandled Rejection (Error): Unknown regularizer: L2. This may be dueto one of the following reasons:

1. The regularizer is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
2. The custom regularizer is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().

model.json 文件内容为

{
 "format": "layers-model",
 "generatedBy": "keras v2.4.0",
 "convertedBy": "TensorFlow.js Converter v2.3.0",
 "modelTopology": {
  "keras_version": "2.4.0",
  "backend": "tensorflow",
  "model_config": {
   "class_name": "Sequential",
   "config": {
    "name": "sequential",
    "layers": [
     {
      "class_name": "InputLayer",
      "config": {
       "batch_input_shape": [null, 22, 13],
       "dtype": "float32",
       "sparse": false,
       "ragged": false,
       "name": "input_dense_input"
      }
     },
     {
      "class_name": "Dense",
      "config": {
       "name": "input_dense",
       "trainable": true,
       "batch_input_shape": [null, 22, 13],
       "dtype": "float32",
       "units": 24,
       "activation": "tanh",
       "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": "GRU",
      "config": {
       "name": "gru",
       "trainable": true,
       "dtype": "float32",
       "return_sequences": true,
       "return_state": false,
       "go_backwards": false,
       "stateful": false,
       "unroll": false,
       "time_major": false,
       "units": 24,
       "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": {
        "class_name": "L2",
        "config": { "l2": 9.999999974752427e-7 }
       },
       "recurrent_regularizer": {
        "class_name": "L2",
        "config": { "l2": 9.999999974752427e-7 }
       },
       "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": false
      }
     },
     {
      "class_name": "Flatten",
      "config": {
       "name": "flatten",
       "trainable": true,
       "dtype": "float32",
       "data_format": "channels_last"
      }
     },
     {
      "class_name": "Dense",
      "config": {
       "name": "dense",
       "trainable": true,
       "dtype": "float32",
       "units": 2,
       "activation": "softmax",
       "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": "categorical_crossentropy",
   "metrics": ["accuracy"],
   "weighted_metrics": null,
   "loss_weights": null,
   "optimizer_config": {
    "class_name": "Nadam",
    "config": {
     "name": "Nadam",
     "learning_rate": 0.0020000000949949026,
     "decay": 0.004000000189989805,
     "beta_1": 0.8999999761581421,
     "beta_2": 0.9990000128746033,
     "epsilon": 1e-7
    }
   }
  }
 },
 "weightsManifest": [
  {
   "paths": ["group1-shard1of1.bin"],
   "weights": [
    { "name": "dense/kernel", "shape": [528, 2], "dtype": "float32" },
    { "name": "dense/bias", "shape": [2], "dtype": "float32" },
    { "name": "gru/gru_cell/kernel", "shape": [24, 72], "dtype": "float32" },
    {
     "name": "gru/gru_cell/recurrent_kernel",
     "shape": [24, 72],
     "dtype": "float32"
    },
    { "name": "gru/gru_cell/bias", "shape": [72], "dtype": "float32" },
    { "name": "input_dense/kernel", "shape": [13, 24], "dtype": "float32" },
    { "name": "input_dense/bias", "shape": [24], "dtype": "float32" }
   ]
  }
 ]
}

Answer 1

选项1
没有类(class)L1和 L2 ;它们只是接口(interface)(更多 here )
有类L1L2它将获取配置并返回正确的正则化器。您可以手动替换所有出现的 L2至L1L2 .
选项 2
注册一个 L2 类

class L2 {

    static className = 'L2';

    constructor(config) {
       return tf.regularizers.l1l2(config)
    }
}
tf.serialization.registerClass(L2);

// now load the model