CODEWITHSUNDEEP
pythonarraysnumpytensorflowkeras
Henrik Hillestad Løvold
Henrik Hillestad Løvold

Reputation: 1301

Wrong shapes for multi input arrays in keras

I'm trying to build a multi input network using keras functional api. Right now I'm stuck as I get the error

ValueError: Error when checking input: expected input_1 to have shape (5,) but got array with shape (1,)

Given my code (below), I don't see how this is possible. The arrays I pass surely have the shapes (5,) and (15,) as my network specifies.

def buildModel():
    # define two sets of inputs
    inputA = ls.Input(shape=(5,))
    inputB = ls.Input(shape=(15,))

    # the first branch operates on the first input
    x = ls.Dense(8, activation="relu")(inputA)
    x = ls.Dense(4, activation="relu")(x)
    x = ks.Model(inputs=inputA, outputs=x)

    # the second branch opreates on the second input
    y = ls.Dense(64, activation="relu")(inputB)
    y = ls.Dense(32, activation="relu")(y)
    y = ls.Dense(4, activation="relu")(y)
    y = ks.Model(inputs=inputB, outputs=y)

    # combine the output of the two branches
    combined = ls.concatenate([x.output, y.output])

    # apply a FC layer and then a regression prediction on the
    # combined outputs
    z1 = ls.Dense(2, activation="relu")(combined)
    z1 = ls.Dense(5, activation="relu")(z1)

    z2 = ls.Dense(2, activation="relu")(combined)
    z2 = ls.Dense(15, activation="relu")(z2)

    # our model will accept the inputs of the two branches and
    # then output a single value
    model = ks.Model(inputs=[x.input, y.input], outputs=[z1, z2])
    model.compile(optimizer='adam', loss='poisson', metrics=['accuracy'])
    return model

def train_model(model, x1, x2, y1, y2):
    print(x1.shape)
    model.fit([x1, x2], [y1, y2], batch_size=1, epochs=100) # Error raised here

if __name__ == "__main__":
    mod = buildModel()
    x1 = np.array([5, 2, 1, 4, 5])
    x2 = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])
    y1 = np.array([0, 0, 1, 0, 0])
    y2 = np.array([0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
    train_model(mod, x1, x2, y1, y2)

The arrays passed are for testing purposes only, the AI will eventually tackle a real problem.

If you think the question isn't well defined or needs more specification, please let me know.

EDIT: Here is the traceback:

Traceback (most recent call last):
  File "/Users/henrihlo/project/ai.py", line 70, in <module>
    train_model(mod, x1, x2, y1, y2)
  File "/Users/henrihlo/project/ai.py", line 49, in train_model
    model.fit([x1, x2], [y1, y2], batch_size=1, epochs=100)
  File "/Users/henrihlo/anaconda3/lib/python3.7/site packages/keras/engine/training.py", line 1154, in fit
    batch_size=batch_size)
  File "/Users/henrihlo/anaconda3/lib/python3.7/site packages/keras/engine/training.py", line 579, in _standardize_user_data
    exception_prefix='input')
  File "/Users/henrihlo/anaconda3/lib/python3.7/site packages/keras/engine/training_utils.py", line 145, in 
standardize_input_data
    str(data_shape))
ValueError: Error when checking input: expected input_1 to have shape (5,) but got array with shape (1,)

Printing the shape of x1 yields (5,).

Upvotes: 1

Views: 105

Answers (1)

Marco Cerliani
Marco Cerliani

Reputation: 22031

you need to feed keras model with 2D array (n_sample, n_feat). A simple reshape in your case does the trick

mod = buildModel()

x1 = np.array([5, 2, 1, 4, 5]).reshape(1,-1) 
x2 = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]).reshape(1,-1)
y1 = np.array([0, 0, 1, 0, 0]).reshape(1,-1)
y2 = np.array([0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]).reshape(1,-1)

train_model(mod, x1, x2, y1, y2)

Upvotes: 1

Related Questions