Optimize TensorFlow & Keras models with L-BFGS from TensorFlow Probability

tf_keras_tfp_lbfgs.py

#! /usr/bin/env python
# -*- coding: utf-8 -*-
# vim:fenc=utf-8
#
# Copyright © 2019 Pi-Yueh Chuang <[email protected]>
#
# Distributed under terms of the MIT license.

"""An example of using tfp.optimizer.lbfgs_minimize to optimize a TensorFlow model.

This code shows a naive way to wrap a tf.keras.Model and optimize it with the L-BFGS
optimizer from TensorFlow Probability.

Python interpreter version: 3.6.9
TensorFlow version: 2.0.0
TensorFlow Probability version: 0.8.0
NumPy version: 1.17.2
Matplotlib version: 3.1.1
"""
import numpy
import tensorflow as tf
import tensorflow_probability as tfp
from matplotlib import pyplot

def function_factory(model, loss, train_x, train_y):
    """A factory to create a function required by tfp.optimizer.lbfgs_minimize.

    Args:
        model [in]: an instance of `tf.keras.Model` or its subclasses.
        loss [in]: a function with signature loss_value = loss(pred_y, true_y).
        train_x [in]: the input part of training data.
        train_y [in]: the output part of training data.

    Returns:
        A function that has a signature of:
            loss_value, gradients = f(model_parameters).
    """

    # obtain the shapes of all trainable parameters in the model
    shapes = tf.shape_n(model.trainable_variables)
    n_tensors = len(shapes)

    # we'll use tf.dynamic_stitch and tf.dynamic_partition later, so we need to
    # prepare required information first
    count = 0
    idx = [] # stitch indices
    part = [] # partition indices

    for i, shape in enumerate(shapes):
        n = numpy.product(shape)
        idx.append(tf.reshape(tf.range(count, count+n, dtype=tf.int32), shape))
        part.extend([i]*n)
        count += n

    part = tf.constant(part)

    @tf.function
    def assign_new_model_parameters(params_1d):
        """A function updating the model's parameters with a 1D tf.Tensor.

        Args:
            params_1d [in]: a 1D tf.Tensor representing the model's trainable parameters.
        """

        params = tf.dynamic_partition(params_1d, part, n_tensors)
        for i, (shape, param) in enumerate(zip(shapes, params)):
            model.trainable_variables[i].assign(tf.reshape(param, shape))

    # now create a function that will be returned by this factory
    @tf.function
    def f(params_1d):
        """A function that can be used by tfp.optimizer.lbfgs_minimize.

        This function is created by function_factory.

        Args:
           params_1d [in]: a 1D tf.Tensor.

        Returns:
            A scalar loss and the gradients w.r.t. the `params_1d`.
        """

        # use GradientTape so that we can calculate the gradient of loss w.r.t. parameters
        with tf.GradientTape() as tape:
            # update the parameters in the model
            assign_new_model_parameters(params_1d)
            # calculate the loss
            loss_value = loss(model(train_x, training=True), train_y)

        # calculate gradients and convert to 1D tf.Tensor
        grads = tape.gradient(loss_value, model.trainable_variables)
        grads = tf.dynamic_stitch(idx, grads)

        # print out iteration & loss
        f.iter.assign_add(1)
        tf.print("Iter:", f.iter, "loss:", loss_value)

        # store loss value so we can retrieve later
        tf.py_function(f.history.append, inp=[loss_value], Tout=[])

        return loss_value, grads

    # store these information as members so we can use them outside the scope
    f.iter = tf.Variable(0)
    f.idx = idx
    f.part = part
    f.shapes = shapes
    f.assign_new_model_parameters = assign_new_model_parameters
    f.history = []

    return f

def plot_helper(inputs, outputs, title, fname):
    """Plot helper"""
    pyplot.figure()
    pyplot.tricontourf(inputs[:, 0], inputs[:, 1], outputs.flatten(), 100)
    pyplot.xlabel("x")
    pyplot.ylabel("y")
    pyplot.title(title)
    pyplot.colorbar()
    pyplot.savefig(fname)

if __name__ == "__main__":

    # use float64 by default
    tf.keras.backend.set_floatx("float64")

    # prepare training data
    x_1d = numpy.linspace(-1., 1., 11)
    x1, x2 = numpy.meshgrid(x_1d, x_1d)
    inps = numpy.stack((x1.flatten(), x2.flatten()), 1)
    outs = numpy.reshape(inps[:, 0]**2+inps[:, 1]**2, (x_1d.size**2, 1))

    # prepare prediction model, loss function, and the function passed to L-BFGS solver
    pred_model = tf.keras.Sequential(
        [tf.keras.Input(shape=[2,]),
         tf.keras.layers.Dense(64, "tanh"),
         tf.keras.layers.Dense(64, "tanh"),
         tf.keras.layers.Dense(1, None)])

    loss_fun = tf.keras.losses.MeanSquaredError()
    func = function_factory(pred_model, loss_fun, inps, outs)

    # convert initial model parameters to a 1D tf.Tensor
    init_params = tf.dynamic_stitch(func.idx, pred_model.trainable_variables)

    # train the model with L-BFGS solver
    results = tfp.optimizer.lbfgs_minimize(
        value_and_gradients_function=func, initial_position=init_params, max_iterations=500)

    # after training, the final optimized parameters are still in results.position
    # so we have to manually put them back to the model
    func.assign_new_model_parameters(results.position)

    # do some prediction
    pred_outs = pred_model.predict(inps)
    err = numpy.abs(pred_outs-outs)
    print("L2-error norm: {}".format(numpy.linalg.norm(err)/numpy.sqrt(11)))

    # plot figures
    plot_helper(inps, outs, "Exact solution", "ext_soln.png")
    plot_helper(inps, pred_outs, "Predicted solution", "pred_soln.png")
    plot_helper(inps, err, "Absolute error", "abs_err.png")
    pyplot.show()

    # print out history
    print("\n"+"="*80)
    print("History")
    print("="*80)
    print(*func.history, sep='\n')

Thanks for the work, it's a pity I only found out now, but I have a doubt. If I use the model created manually by tf instead of the model created by tf.keras.Sequentia, can I continue to use the above code?

I think so, but I haven't been using TF for more than 3 years, so I can't say for sure.

@piyueh
Thank you for your quick reply.

I'm not sure if I had a specific problem with my implementation or TF version, but I had to modify the line number 40

shapes = tf.shape_n(model.trainable_variables)

with this

shapes = []
for i in range(len(model.trainable_variables)):
    shapes.append(model.trainable_variables[i].shape)

I'm using a tf.keras.Sequential() model.

Thank you very much!!!

Hi, thank you for providing the script! It generally works for me, but I encounter issues when using a pre-trained model. When I insert pred_model.set_weights(keras_model.get_weights()) at line 140, where keras_model is my pre-trained model, no changes or optimizations occur. The pre-trained keras_model is returned with the exact same weights I provided. Do you know what might be causing this?

Additionally, your method should work for multiple outputs, correct?

Hi @Stefanos1100 thank for the comment. Unfortunately, I haven't been using TensorFlow and haven't been doing machine/deep learning problems for several years. I may not be able to look into the question at this moment.