[FIXED] InvalidArgumentError: cannot compute MatMul as input #1(zero-based) was expected to be a double tensor but is a float tensor [Op:MatMul] name:

Issue

The data type of input_data is an array of numpy.float64 but the code still fails inside a tensorflow library because it's not a 'double'. Not sure how to remedy this.

import tensorflow as tf
import numpy as np

input_data = np.random.uniform(low=0.0, high=1.0, size=100)
print("type(input_data):", type(input_data), "type(input_data[0]):", type(input_data[0]))

class ArtificialNeuron(tf.Module):
    def __init__(self):
        self.w = tf.Variable(tf.random.normal(shape=(1, 1)))
        self.b = tf.Variable(tf.zeros(shape=(1,)))

    def __call__(self, x):
        return tf.sigmoid(tf.matmul(x, self.w) + self.b)

neuron = ArtificialNeuron()

# Fails here: InvalidArgumentError: cannot compute MatMul as input #1(zero-based) was expected to be a double tensor but is a float tensor [Op:MatMul] name: 
output_data = neuron(input_data)  

Solution

The error is thrown because you are mixing float32 and float64 tensors without conversion. By default numpy uses float64 while TensorFlow uses float32. Normally the higher level modules perform the conversion, but I think you are using low-level building blocks, so you have to do the conversion yourself.

You can simply test this:

import numpy as np
import tensorflow as tf

x = np.array([[1.0]], dtype=np.float64)
w = tf.zeros(shape=(1,1))
tf.matmul(x, w)

# => InvalidArgumentError: cannot compute MatMul as input #1(zero-based) was expected to be a double tensor but is a float tensor [Op:MatMul]
# Changing np.float32 to np.float64 above the code works...

So you either have to convert your input to float32 by using np.float32(input_data), or use everywhere float64 tensors. You might also change the default precision of TensorFlow to float64 as described here: TensorFlow default precision mode?

In short, to fix your code, replace input_data in the last line by np.float32(input_data) and size=100 in the definition of input_data by size=(100,1).

Answered By - gabalz