CODEWITHSUNDEEP
pythonkerastensordot-product
xyin
xyin

Reputation: 417

tensor dot product in keras

I am new to keras, and got some problems understanding the keras.layers.Dot() layer.

I am trying to calculate a dot product of two vectors.

from keras.layers import Input, Dot
from keras.models import Model
import numpy as np

x1 = Input(shape=(4,))
x2 = Input(shape=(4,))
y1 = Dot(axes=1)([x1,x2])
model = Model(inputs=[x1, x2], outputs=y1)

a1 = np.arange(4)
a2=np.arange(4)
model.predict([a1,a2])

I expect the output to be 14=0+1^2+2^2+3^2. However, I got error message like below:

ValueError: Error when checking input: expected input_46 to have shape (4,) but got array with shape (1,)

I tried to run model.get_config(), and below is the corresponding information about the graph of the model. As you can see input_46 is x1, and input_47 is x2.

{'name': 'model_19',
 'layers': [{'name': 'input_46',
   'class_name': 'InputLayer',
   'config': {'batch_input_shape': (None, 4),
    'dtype': 'float32',
    'sparse': False,
    'name': 'input_46'},
   'inbound_nodes': []},
  {'name': 'input_47',
   'class_name': 'InputLayer',
   'config': {'batch_input_shape': (None, 4),
    'dtype': 'float32',
    'sparse': False,
    'name': 'input_47'},
   'inbound_nodes': []},
  {'name': 'dot_20',
   'class_name': 'Dot',
   'config': {'name': 'dot_20',
    'trainable': True,
    'axes': 1,
    'normalize': False},
   'inbound_nodes': [[['input_46', 0, 0, {}], ['input_47', 0, 0, {}]]]}],
 'input_layers': [['input_46', 0, 0], ['input_47', 0, 0]],
 'output_layers': [['dot_20', 0, 0]]}

Is there anything I didn't do right? Thanks!

UPDATE

The following code worked:

x1 = Input(shape=(4,))
x2 = Input(shape=(4,))
y1 = Dot(axes=1)([x1,x2])
model = Model(inputs=[x1, x2], outputs=y1)
a1 = np.arange(4).reshape(1,4)
a2=np.arange(4).reshape(1,4)
model.predict([a1,a2])

or

from keras.layers import Input, Dot
from keras.models import Model
import numpy as np

x1 = Input(shape=(4,))
x2 = Input(shape=(4,))
y1 = Dot(axes=1)([x1,x2])
model = Model(inputs=[x1, x2], outputs=y1)

a1 = np.arange(4)
a2=np.arange(4)
model.predict([[a1],[a2]])

Upvotes: 2

Views: 925

Answers (1)

alift
alift

Reputation: 1928

Keras waits for get a batch of your inputs. like a vector of size N*4 if you have N inputs of the dimension 4. So if you want to send a single vector, it shoul dbe in the dimension of 1*4 and not 4. So you have two choices:

  1. Change the

    a1 = np.arange(4) a2=np.arange(4)

To

a1 = np.arange(1,4)
a2=np.arange(1,4)
  1. or Try model.predict([[a1],[a2]]) . Should solve your issue.

Upvotes: 1

Related Questions