Valueerror: attempt to convert a value (probs) with an unsupported type to a tensor

PennyLane Help
1 
import tensorflow as tf
import pennylane as qml
from pennylane import numpy as np


def qnn_patch(image, weights):
   
    dev = qml.device("default.qubit.tf", wires=9)

    # Encode the input image into the quantum circuit using patches of 3x3 pixels
    for i in range(3):
        for j in range(3):
            qml.templates.embeddings.AngleEmbedding(image[i:i+3, j:j+3], wires=list(range(3*(i*3+j), 3*(i*3+j+1))))

    # Apply the trainable weights to each qubit
    for i in range(9):
        qml.RY(weights[i], wires=i)

   
    return qml.probs(wires=range(9))


def cost(weights, images, labels):
    predictions = [qnn_patch(image, weights) for image in images]
    loss = tf.keras.losses.sparse_categorical_crossentropy(labels, predictions)
    return tf.reduce_mean(loss)


(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

# Preprocess the data
x_train = x_train.astype(np.float32) / 255.0
x_test = x_test.astype(np.float32) / 255.0
y_train = y_train.astype(np.int64)
y_test = y_test.astype(np.int64)


opt = tf.keras.optimizers.Adam(learning_rate=0.1)
weights = tf.Variable(np.random.uniform(low=-np.pi/2, high=np.pi/2, size=(9)), dtype=np.float32)


for epoch in range(10):
    with tf.GradientTape() as tape:
        loss = cost(weights, x_train[:100], y_train[:100])
    gradients = tape.gradient(loss, [weights])
    opt.apply_gradients(zip(gradients, [weights]))
    print("Epoch {}: Loss = {}".format(epoch+1, loss))

# Evaluate 
predictions = [np.argmax(qnn_patch(image, weights)) for image in x_test]
accuracy = np.mean(predictions == y_test)
print("Accuracy: {}".format(accuracy))

During training, it throws an error of valueerror., as its returns list of probabilities from each wires.

2

Hey @Amandeep,

As your error suggests, there’s something wrong with what your quantum function qnn_patch is outputting in that it hasn’t been wrapped as a QNode. You need to create the device beforehand and decorate with @qml.qnode. I fixed that, but it seems that there’s still an error with your angle embedding procedure. NB: I printed list(range(3*(i*3+j), 3*(i*3+j+1))):

import tensorflow as tf
import pennylane as qml
from pennylane import numpy as np

dev = qml.device("default.qubit.tf", wires=9)

@qml.qnode(dev)
def qnn_patch(image, weights):
    # Encode the input image into the quantum circuit using patches of 3x3 pixels
    for i in range(3):
        for j in range(3):
            print(list(range(3*(i*3+j), 3*(i*3+j+1))))
            qml.templates.embeddings.AngleEmbedding(image[i:i+3, j:j+3], wires=list(range(3*(i*3+j), 3*(i*3+j+1))))

    # Apply the trainable weights to each qubit
    for i in range(9):
        qml.RY(weights[i], wires=i)

    return qml.probs(wires=range(9))


def cost(weights, images, labels):
    predictions = [qnn_patch(image, weights) for image in images]
    loss = tf.keras.losses.sparse_categorical_crossentropy(labels, predictions)
    return tf.reduce_mean(loss)


(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

# Preprocess the data
x_train = x_train.astype(np.float32) / 255.0
x_test = x_test.astype(np.float32) / 255.0
y_train = y_train.astype(np.int64)
y_test = y_test.astype(np.int64)


opt = tf.keras.optimizers.Adam(learning_rate=0.1)
weights = tf.Variable(np.random.uniform(low=-np.pi/2, high=np.pi/2, size=(9,)), dtype=np.float32)


for epoch in range(10):
    with tf.GradientTape() as tape:
        loss = cost(weights, x_train[:100], y_train[:100])
    gradients = tape.gradient(loss, [weights])
    opt.apply_gradients(zip(gradients, [weights]))
    print("Epoch {}: Loss = {}".format(epoch+1, loss))

# Evaluate 
predictions = [np.argmax(qnn_patch(image, weights)) for image in x_test]
accuracy = np.mean(predictions == y_test)
print("Accuracy: {}".format(accuracy))

'''
[0, 1, 2]
[3, 4, 5]
[6, 7, 8]
[9, 10, 11]
[12, 13, 14]
[15, 16, 17]
[18, 19, 20]
[21, 22, 23]
[24, 25, 26]

WireError: Did not find some of the wires (9,) on device with wires (0, 1, 2, 3, 4, 5, 6, 7, 8).
'''

You are putting gates on wires that don’t exist. Hope this helps!