I'm learning about Gaussian processes and I'm using GPflow to do some exercises and tests.

I was studying the notebook of Stochastic Variational Inference for scalability with SVGP (.4.0/notebooks/advanced/gps_for_big_data.html) and I have managed to replicate most of it, but I have found a bug associated with the use of the Adam optimizer. Attached is the block of code that causes the error for clarity

minibatch_size = 100

# We turn off training for inducing point locations
gpflow.set_trainable(m.inducing_variable, False)


def run_adam(model, iterations):
    """
    Utility function running the Adam optimizer

    :param model: GPflow model
    :param interations: number of iterations
    """
    # Create an Adam Optimizer action
    logf = []
    train_iter = iter(train_dataset.batch(minibatch_size))
    training_loss = model.training_loss_closure(train_iter, compile=True)
    optimizer = tf.optimizers.Adam()

    @tf.function
    def optimization_step():
        optimizer.minimize(training_loss, model.trainable_variables)

    for step in range(iterations):
        optimization_step()
        if step % 10 == 0:
            elbo = -training_loss().numpy()
            logf.append(elbo)
    return logf

And the error is as follows:

AttributeError: in user code:

File "C:\Users\D\AppData\Local\Temp\ipykernel_16304\2914892647.py", line 16, in optimization_step  *
    optimizer.minimize(training_loss, model.trainable_variables)

AttributeError: 'Adam' object has no attribute 'minimize'

I don't really know how to solve this problem, because I'm just starting to learn how to use TensorFlow.

P.S. I'm using GPflow 2.4.0 and Tensor Flow 2.8.1

Thank you very much for your time

I'm learning about Gaussian processes and I'm using GPflow to do some exercises and tests.

I was studying the notebook of Stochastic Variational Inference for scalability with SVGP (https://gpflow.github.io/GPflow/2.4.0/notebooks/advanced/gps_for_big_data.html) and I have managed to replicate most of it, but I have found a bug associated with the use of the Adam optimizer. Attached is the block of code that causes the error for clarity

minibatch_size = 100

# We turn off training for inducing point locations
gpflow.set_trainable(m.inducing_variable, False)


def run_adam(model, iterations):
    """
    Utility function running the Adam optimizer

    :param model: GPflow model
    :param interations: number of iterations
    """
    # Create an Adam Optimizer action
    logf = []
    train_iter = iter(train_dataset.batch(minibatch_size))
    training_loss = model.training_loss_closure(train_iter, compile=True)
    optimizer = tf.optimizers.Adam()

    @tf.function
    def optimization_step():
        optimizer.minimize(training_loss, model.trainable_variables)

    for step in range(iterations):
        optimization_step()
        if step % 10 == 0:
            elbo = -training_loss().numpy()
            logf.append(elbo)
    return logf

And the error is as follows:

AttributeError: in user code:

File "C:\Users\D\AppData\Local\Temp\ipykernel_16304\2914892647.py", line 16, in optimization_step  *
    optimizer.minimize(training_loss, model.trainable_variables)

AttributeError: 'Adam' object has no attribute 'minimize'

I don't really know how to solve this problem, because I'm just starting to learn how to use TensorFlow.

P.S. I'm using GPflow 2.4.0 and Tensor Flow 2.8.1

Thank you very much for your time

• python
• tensorflow
• gpflow

1 Answer 1

It looks like I finally found the solution.

The problem was related to the version of TensorFlow. Using the versions recommended on the installation page there is no problem.

https://gpflow.github.io/GPflow/develop/installation.html

I tried to solve the example with gpflow 2.9.1 and tensorflow 2.12.0 and everything worked fine.

Best regards