Keras 模型输入：3 个数组与 3 个张量的元组不一样？

Question

I am trying to fit my inputs to the keras model I have prepared. The input layers of my network are:

path_source_token_input = Input(shape=(MAX_CONTEXTS,), dtype=tf.int32)
path_input = Input(shape=(MAX_CONTEXTS,), dtype=tf.int32)
path_target_token_input = Input(shape=(MAX_CONTEXTS,), dtype=tf.int32)

And I specify the input this way:

inputs = (path_source_token_input, path_input, path_target_token_input)
model = tf.keras.Model(inputs=inputs, outputs=learned)  # outputs not important at this point

Then I load my data from a csv file, do the appropriate preprocessing and create a dataset object which looks like this in debug:

Now my model compiles, all is good and then I try to fit it to the data:

model_x.compile(loss='mse', optimizer='adam', metrics=['accuracy'])
history = model_x.fit(context_paths, epochs=20, verbose=2)

But it throws this error:

ValueError: Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 3 array(s), for inputs ['input_1', 'input_2', 'input_3'] but instead got the following list of 1 arrays: []...

At this point I am not sure what is wrong, because in debug it seems that my dataset is a tuple of length 3 (the way I want it and specify as 'input') but something goes wrong. I would appreciate any help, thank you.

Answer 1

You built the model to get three different inputs: (path_source_token_input, path_input, path_target_token_input) .

You need data that is a list of 3 arrays. One array for path_source_token_input , another array for path_input , and a third array for path_target_token_input .

context_paths = [array1, array2, array3] .

Where:

array1.shape == (anything, MAX_CONTEXTS)
array2.shape == (anything, MAX_CONTEXTS)   
array3.shape == (anything, MAX_CONTEXTS)

Don't think that the outputs are not important, you cannot fit without outputs if you didn't prepare a model with a special loss for this.

---

Data process

Just create the arrays, don't use a dataset:

def loadFile(filename):

    with open(filename, 'r') as file:
        lines = file.readlines()
                

    triplets = [l.split(" ") for l in lines]  #(16000, 430)
    singles = [[t.split(',') for t in line] for line in triplets]  #(16000, 430, 3)

    data = np.array(singles).astype(np.int32)
    data_source = data[:,:,0]
    data_path = data[:,:,1]
    data_target = data[:,:,2]    

    return [data_source, data_path, data_target]