I was experimenting with different shapes of input array to the Keras sequential layer and the input_dim property. Like below:
X_train = np.array([0., 1, 2, 3, 4, 5])
print(X_train.shape)
X_train_reshap = X_train.reshape(-1,1)
print(X_train_reshap.shape)
model = Sequential(
[
tf.keras.layers.Dense(1, input_dim=1, activation = ‘sigmoid’, name=‘L1’)
]
)
layer = model.get_layer(‘L1’)
model.summary()
a1 = model.predict(X_train_reshap)
print(a1)
Here even though the value for input_dim is 1. the model is working for both X_train & X_train_reshap.
The shape of X_train is (6,) which is a 1D array as it is a flat array and the shape of X_train_reshap is (6, 1) which is a 2D array as it has 2 dimensions i.e. 6 Rows & 1 Column. How it’s working? though it’s emitting different results for a1.
As per my understanding the input_dim should have been 2 if I am using X_train_reshap as input but when I am doing that I am having the below error:
Input 0 of layer “L1” is incompatible with the layer: expected axis -1 of input shape to have value 2, but received input with shape (None, 1) .
So I believe I my understanding is not right either about 1D & 2D array or about the input_dim argument in the keras API.
Can anyone please explain?
