Gaussian Process example in tensorflow website is giving error?

tonygrey · June 28, 2021, 6:21pm

The Gaussian process regression example (Gaussian Process Regression in TensorFlow Probability) is giving following error

ValueError: No gradients provided for any variable: ['amplitude:0', 'length_scale:0', 'observation_noise_variance_var:0'].

This error goes away when I comment out the tf.function decorator.
I am using tensorflow 2.5.0.
What is the cause of the error?

lgusm · June 29, 2021, 8:47am

Thanks for pointing that out @tonygrey

I filled a bug on that and I’ll keep you posted!

Bhack · June 29, 2021, 1:38pm

There was already a ticket:

github.com/tensorflow/probability

Gaussian Process Regression in TensorFlow Probability Notebook not working

opened 12:08PM - 14 Jun 21 UTC

jung-benjamin

When running the tutorial notebook *Gaussian Process Regression in TensorFlow Pr…obability* in Google Colab I run into the following issue: I execute the code cells in order, without changing any code. The cell containing the following code produces an error. ``` # Now we optimize the model parameters. num_iters = 1000 optimizer = tf.optimizers.Adam(learning_rate=.01) # Store the likelihood values during training, so we can plot the progress lls_ = np.zeros(num_iters, np.float64) for i in range(num_iters): with tf.GradientTape() as tape: loss = -target_log_prob(amplitude_var, length_scale_var, observation_noise_variance_var) grads = tape.gradient(loss, trainable_variables) optimizer.apply_gradients(zip(grads, trainable_variables)) lls_[i] = loss print('Trained parameters:') print('amplitude: {}'.format(amplitude_var._value().numpy())) print('length_scale: {}'.format(length_scale_var._value().numpy())) print('observation_noise_variance: {}'.format(observation_noise_variance_var._value().numpy())) ``` The error message is: ``` ValueError Traceback (most recent call last) <ipython-input-81-0b42fdbef836> in <module>() 10 observation_noise_variance_var) 11 grads = tape.gradient(loss, trainable_variables) ---> 12 optimizer.apply_gradients(zip(grads, trainable_variables)) 13 lls_[i] = loss 14 1 frames /usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/optimizer_v2/utils.py in filter_empty_gradients(grads_and_vars) 74 if not filtered: 75 raise ValueError("No gradients provided for any variable: %s." % ---> 76 ([v.name for _, v in grads_and_vars],)) 77 if vars_with_empty_grads: 78 logging.warning( ValueError: No gradients provided for any variable: ['amplitude:0', 'length_scale:0', 'observation_noise_variance_var:0']. ``` After consulting Stack Overflow I tried adding `tape.watch(trainable_variables)`, but this did not change anything. Might someone be able to help figure out why this is happening?

Bhack · June 29, 2021, 1:42pm

I suppose that the main problem is that this and other notebooks are not under CI testing in the TFP repo.

lgusm · June 23, 2022, 9:34am

Apparently this is fixed.
maybe it was something with libraries versioning