Keras: Input 0 of layer sequential is incompatible with the layer

Question

I am trying to create a neural network model with one hidden layer and then trying to evaluate it, but I am getting an error that I am not able to understand clearly:

ValueError: Input 0 of layer sequential_1 is incompatible with the layer: : expected min_ndim=2, found ndim=1. Full shape received: [30]

It looks like I have an error with the dimensions of my input layer, but I can't quite spot what. I've googled and looked on stackoverflow, but haven't found anything that worked so far. Any help please?

Here's a minimal working example:

import tensorflow as tf

# Define Sequential model with 3 layers
input_dim = 30
num_neurons = 10
output_dim = 12
model = tf.keras.Sequential(
    [
        tf.keras.layers.Dense(input_dim, activation="relu", name="layer1"),
        tf.keras.layers.Dense(num_neurons, activation="relu", name="layer2"),
        tf.keras.layers.Dense(output_dim, name="layer3"),
    ]
)
model(tf.ones(input_dim))

piterbarg · Accepted Answer · 2020-11-09 00:20:51Z

2

Layers have an input and output dimension. For layers in the "middle" of the NN, they figure out their input domain from the output domain of the previous layer. The only exception is the first layer that has nothing to go by, and requires input_dim to be set. Here is how to fix your code. Note how we pass the dimensions. First (hidden) layer is input_dim x num_neurons, second (output layer) num_neurons x output_dim

You can stick more layers in between the two; they only require the first argument, their output dimension

also note I had to fix your last line as well, tf.ones needs to be a 2D shape num_observation x input_dim

import tensorflow as tf

# Define Sequential model with 1 hidden layer
input_dim = 30
num_neurons = 10
output_dim = 12
model = tf.keras.Sequential(
    [
        tf.keras.layers.Dense(num_neurons, input_dim = input_dim, activation="relu", name="layer1"),
        tf.keras.layers.Dense(output_dim, name="layer3"),
    ]
)
model(tf.ones((1,input_dim)))

produces (for me; I think the numbers are essentially random initialization)

<tf.Tensor: shape=(1, 12), dtype=float32, numpy=
array([[ 0.06973769, -0.1798143 , -0.2920275 ,  0.84811246,  0.44899416,
        -0.10300556,  0.00831143, -0.16158538,  0.13395026,  0.4352504 ,
         0.19114715,  0.44100884]], dtype=float32)>

answered Nov 9, 2020 at 0:20

piterbarg

8,2292 gold badges9 silver badges22 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Shruti Over a year ago

Thank you! I see that there were more than one problems with my code, thank you for helping to fix them and for the clear explanation!

Collectives™ on Stack Overflow

Keras: Input 0 of layer sequential is incompatible with the layer

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related