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

**URL:** <https://discuss.pennylane.ai/t/valueerror-attempt-to-convert-a-value-probs-with-an-unsupported-type-to-a-tensor/2877>\
**Category:** PennyLane Help\
**Created:** [April 24, 2023, 3:49pm UTC](https://discuss.pennylane.ai/t/valueerror-attempt-to-convert-a-value-probs-with-an-unsupported-type-to-a-tensor/2877 "2023-04-24T15:49:29Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Amandeep](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/amandeep/32/385_2.png) [@Amandeep](https://discuss.pennylane.ai/u/Amandeep)\
**Post date:** [April 24, 2023, 3:49pm UTC](https://discuss.pennylane.ai/t/valueerror-attempt-to-convert-a-value-probs-with-an-unsupported-type-to-a-tensor/2877/1 "2023-04-24T15:49:29Z")

</div>

```python
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.

---

<div class="post-metadata">

**Author:** ![isaacdevlugt](https://yyz2.discourse-cdn.com/flex012/user_avatar/discuss.pennylane.ai/isaacdevlugt/32/1159_2.png) [@isaacdevlugt](https://discuss.pennylane.ai/u/isaacdevlugt)\
**Post date:** [April 25, 2023, 3:35pm UTC](https://discuss.pennylane.ai/t/valueerror-attempt-to-convert-a-value-probs-with-an-unsupported-type-to-a-tensor/2877/2 "2023-04-25T15:35:07Z")

</div>

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)))`:

```python
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!