[FIXED] How to solve "Input 0 of layer conv1d is incompatible with the layer:" error?

Issue

Please can you help me understand the error I am getting from the model I am trying to build?

I have a training, validation and test set. The training data has the following shape:

input_shape = train.shape[1:] #(1500,)

I have written the following model using Keras:

input = Input(shape=(input_shape))

# Conv1D + global max pooling
x = layers.Conv1D(filters=32, padding="valid", activation="relu", strides=1, kernel_size=4)(input)
x = layers.Conv1D(filters=32, padding="valid", activation="relu", strides=1, kernel_size=4)(x)
x = layers.GlobalMaxPooling1D()(x)

x = layers.Dense(128, activation="relu")(x)
x = layers.Dropout(0.5)(x)

predictions = layers.Dense(1,kernel_initializer='normal', name="predictions")(x)

model = tf.keras.Model(input, predictions)

model.compile(loss="mean squared error", optimizer="adam", metrics=[concordance_index])

I get the following error:

ValueError                                Traceback (most recent call last)
<ipython-input-60-59c3578104d3> in <module>()
      6 
      7 # Conv1D + global max pooling
----> 8 x = layers.Conv1D(filters=32, padding="valid", activation="relu", strides=1, kernel_size=4)(protein_input)
      9 x = layers.Conv1D(filters=32, padding="valid", activation="relu", strides=1, kernel_size=4)(x)
     10 x = layers.GlobalMaxPooling1D()(x)

5 frames
/usr/local/lib/python3.7/dist-packages/keras/engine/input_spec.py in assert_input_compatibility(input_spec, inputs, layer_name)
    230                          ', found ndim=' + str(ndim) +
    231                          '. Full shape received: ' +
--> 232                          str(tuple(shape)))
    233     # Check dtype.
    234     if spec.dtype is not None:

ValueError: Input 0 of layer conv1d_49 is incompatible with the layer: : expected min_ndim=3, found ndim=2. Full shape received: (None, 1500)

Is my Input layer incorrect? Or is it due to the ordering of Conv1d, MaxPooling layers?

Solution

Since a Conv1D layer expects input as batch_shape + (steps, input_dim), you need to add a new dimension. So:

X = tf.expand_dims(X,axis=2)
print(X.shape)     # X.shape=(Samples, 1500, 1) 

Then, you have X shape as (Samples,150,1)

Now, let's specify the input shape as:

input = Input(shape=(X.shape[1:])  

Answered By - Kaveh