I built and trained a model with Keras, and saved it with tensorflowjs converter (the tfjs.converters.save_keras_model() function).
Later, when trying to load it in tensorflowjs, I get the following error:
Error: The first layer in a Sequential model must get an 'inputShape' or 'batchInputShape' argument.
But after checking in the JSON file containing the model structure, there IS an input shape specified.
Any idea why tfjs cannot load it?
Could it be caused by the different variable names (batch_input_shape in my JSON file and batchInputShape in the error message).
Here is how I build and train the model :
model.add(LSTM(128, dropout=0.2, input_shape=(time_steps, input_dim) ))
model.add(Dense(output_dim, activation='sigmoid'))
model.compile(optimizer='adam', loss='mse', metrics=['accuracy'])
model.fit_generator(generator=train_generator,
steps_per_epoch=steps_per_epoch,
epochs=epochs,
validation_data=valid_generator,
validation_steps=valid_steps
)
Here is the JSON file. I don't know where the third null variable comes from, but if I change it I get an error saying the number of dimensions is wrong.
"format": "layers-model",
"generatedBy": "keras v2.3.1",
"convertedBy": "TensorFlow.js Converter v1.4.0",
"modelTopology": {
"keras_version": "2.3.1",
"backend": "tensorflow",
"model_config": {
"class_name": "Sequential",
"config": {
"name": "sequential_1",
"layers": [
{
"class_name": "LSTM",
"config": {
"name": "lstm_1",
"trainable": true,
"batch_input_shape": [null, 10, 100],
"dtype": "float32",
"return_sequences": false,
"return_state": false,
"go_backwards": false,
"stateful": false,
"unroll": false,
"units": 128,
"activation": "tanh",
"recurrent_activation": "sigmoid",
"use_bias": true,
"kernel_initializer": {
"class_name": "VarianceScaling",
"config": {
"scale": 1.0,
"mode": "fan_avg",
"distribution": "uniform",
"seed": null
}
},
"recurrent_initializer": {"class_name": "Orthogonal", "config": {"gain": 1.0, "seed": null}},
"bias_initializer": {"class_name": "Zeros", "config": {}},
"unit_forget_bias": true,
"kernel_regularizer": null,
"recurrent_regularizer": null,
"bias_regularizer": null,
"activity_regularizer": null,
"kernel_constraint": null,
"recurrent_constraint": null,
"bias_constraint": null,
"dropout": 0.2,
"recurrent_dropout": 0.0,
"implementation": 2
}
},
{"class_name": "Dense", "config": {"name": "dense_1", "trainable": true, "dtype": "float32", "units": 29, "activation": "sigmoid", "use_bias": true, "kernel_initializer":
{"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "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": {
"optimizer_config": {
"class_name": "Adam",
"config": {
"learning_rate": 0.0010000000474974513,
"beta_1": 0.8999999761581421,
"beta_2": 0.9990000128746033,
"decay": 0.0,
"epsilon": 1e-07,
"amsgrad": false
}
},
"loss": "mse",
"metrics": ["accuracy"],
"weighted_metrics": null,
"sample_weight_mode": null,
"loss_weights": null
}
},
"weightsManifest": [{
"paths": ["group1-shard1of1.bin"],
"weights": [
{"name": "dense_1/kernel", "shape": [128, 29], "dtype": "float32"},
{"name": "dense_1/bias", "shape": [29], "dtype": "float32"},
{"name": "lstm_1/kernel", "shape": [100, 512], "dtype": "float32"},
{"name": "lstm_1/recurrent_kernel", "shape": [128, 512], "dtype": "float32"},
{"name": "lstm_1/bias", "shape": [512], "dtype": "float32"}
]
}]
}
inputShape, can you please addinput_shape": [10, 100]and see what error you have ? - edkevekedError: Input 0 is incompatible with layer lstm_2: expected ndim=3, found ndim=2- Seldiinput_shape": [null, 10, 100], do you still have the same issue ? - edkevekedError: The first layer in a Sequential model must get an 'inputShape' or 'batchInputShape' argument.like previously - Seldiinput_shape": [10, 100]I do not have the same error message as I did the first time - Seldi