CODEWITHSUNDEEP
pythonmachine-learningkerastraining-datakeras-layer
julieb
julieb

Reputation: 43

Input shape and Keras

I have a set of training data where each input is a vector of length 138. I have 519 of these vectors for a batch size of 519. These are not images, just real-valued numbers.

I'm trying to start with a 2 layer dense Keras model:

model = keras.Sequential([
    layers.Dense(200, activation=tf.nn.relu, input_shape=[138]),
    layers.Dense(200, activation=tf.nn.relu),
    layers.Dense(1)
])

When I build the model I get the following error:

Error when checking input: expected dense_27_input to have shape (138,) but got array with shape (519,).

Where in Keras do I distinguish batch size from number of input features? layers.Dense() seems to assume that my input is in rows vs. columns.

Upvotes: 0

Views: 3311

Answers (1)

today
today

Reputation: 33470

Keras expects the first axis to be the batch axis. Therefore if you have 519 training samples where each one is a vector of length 138, the array you pass to the fit method must have a shape of (519, 138). So if currently the array of training data has a shape of (138, 519), simply transpose it to make the shape consistent:

import numpy as np

train_data = np.transpose(train_data)

Upvotes: 1

Related Questions