pythontensorflowmachine-learningtensorflow2.0

tensorflow keras Model.fit returning: ValueError: Unrecognized data type

I'm trying to train a keras model with 2 inputs: an image part that's a tf.data.Dataset and a nor mal part represented by a pd.DataFrame

from tensorflow.keras.optimizers import Adam
opt = Adam(learning_rate=1e-3, decay=1e-3 / 200)

model.compile(loss="mean_absolute_percentage_error", optimizer=opt)

model.fit(
    x=[df.loc[:, df.columns != 'target'], ds.batch(8)], y=df["target"],
    epochs=200)

I was trying to fit the model but I get ValueError

ValueError: Unrecognized data type: x=[...][401059 rows x 52 columns]
, <_BatchDataset element_spec=(TensorSpec(shape=(None, 32, 256, 256, 3), 
dtype=tf.float32, name=None), 
TensorSpec(shape=(None, 32, 256, 256, 3), dtype=tf.float32, name=None))>] (of type <class 'list'>)
Solution

  • the problem was an error in tensoflow zipping and reformating dataset helped

    def post_zip_process(example1, example2):
    reshaped_input = tf.transpose(example1[0], [0, 1, 2 ,-1])
    reshaped_input = reshaped_input[0, :, :, :]
    print(reshaped_input.shape)
    return (reshaped_input, example2[0]), example1[1]