"WARNING:tensorflow: You are casting an input of type complex64 to an incompatible dtype float32. This will discard the imaginary part and may not be what you intended." about cvnn HOT 5 OPEN

remove dype or change it to tf.complex64

eg:
import tensorflow as tf
tf. device("gpu:1")
model = tf.keras.models.Sequential()
model.add(complex_layers.ComplexInput(input_shape=input_shape))
model.add(complex_layers.ComplexConv2D(32, (2, 2), activation='cart_leaky_relu', padding='same'))
model.add(complex_layers.ComplexMaxPooling2D((2, 2)))
model.add(complex_layers.ComplexConv2D(64, (2, 2), activation='cart_leaky_relu', padding='same'))
model.add(complex_layers.ComplexMaxPooling2D((2, 2)))
model.add(complex_layers.ComplexConv2D(128, (2, 2), activation='cart_leaky_relu', padding='same'))
model.add(complex_layers.ComplexMaxPooling2D((2, 2)))
model.add(complex_layers.ComplexConv2D(256, (2, 2), activation='cart_leaky_relu', padding='same'))
model.add(complex_layers.ComplexMaxPooling2D((2, 2)))
model.add(complex_layers.ComplexFlatten())

from cvnn.

@nipulseervi Thank you for your previous help. I've successfully removed the dtype and modified the activations to "cart_leaky_relu", it removed the warning. However, I'm wondering if this "cvnn" library other activation functions can handle complex numbers directly without considering the real and imaginary parts separately.
I've tried using activation functions like modrelu, crelu, and zrelu, but I keep receiving the following warning:
“WARNING:tensorflow:You are casting an input of type complex64 to an incompatible dtype float32. This will discard the imaginary part and may not be what you intended.”
I would like to know if these activations can handle complex numbers directly, preserving both the real and imaginary parts. I would appreciate any guidance on how to use these models.

from cvnn.

This solution might not be efficient but it solved the error for 'modrelu'.
Within cvnn/activations.py change the function modrelu to:
"
def modrelu(z: Tensor, b: float = 1., c: float = 1e-3) -> Tensor:
abs_z = tf.math.abs(z)
mod_relu = tf.keras.activations.relu(abs_z + b)
return tf.complex(mod_relu * tf.math.real(z) / (abs_z + c), mod_relu * tf.math.imag(z) / (abs_z + c))
"

from cvnn.

Sorry for my late reply.
This error was because of your line that uses 'relu' activation function. Here you are using Tensorflow ReLU. Tensorflow is not prepared to deal with complex values so it cast it to float. My library then realizes this happens and because I cannot go to Tensorflow to change stuff I at least warn you about this. Please use only activation functions from cvnn toolbox or code an activation function yourself.

from cvnn.