Machine Learning Tensor Flow Examples
https://github.com/adrianbevan/TensorFlow-Tutorial
Copyright (C) 2020 Adrian Bevan, Queen Mary University of London
This code provides a set of TensorFlow and Keras tutorials that covers the following machine learning problems:
-
Linear Regression
- LinearRegression.py: Explore the problem of a least squares optimisation, fitting to a straight line (y-mx+c), where in this case the uncertainty on the data are set to be a relative percentage of the value of the input data.
-
MultiLayer Perceptrons (MLP)
-
NN.py Fit a 2 layer MLP to the MNIST data.
-
NN_parabola.py 2 layer MLP fitting a parabolic function (y=x^2)
-
-
Convolutional Neural Networks (CNNs)
- CNN.py: CNN fitting MNIST or CFAR10 data. To fit the MNIST data ensure that UseMNIST = True is set. In order to fit CFAR10 data then set UseMNIST to False.
To get started run the following command:
source setup.sh
This will create directories required for output figures and log files, and prints a brief summary of the hyperparameter scan scripts.
Each of the examples has been written to allow the user to explore pedegogical aspects of machine learning, starting with probing the performance of optimisation via the linear regression example, through to training performance via test vs train loss function and accuracy as a function of the training epochs.
These examples are accompanied by a set of notes that indicate suggested exercises using these examples in order to build a deeper understanding of the methods, their pathologies and some ways that will allow users to get hands on experience of some of the basics related to machine learining.
If you find these useful then you might be interested in looking at some of my machine learning-related teaching materials that can be found online at:
https://pprc.qmul.ac.uk/~bevan/teaching.html
Note that requirements.txt lists specific package versions that this tutorial has been written with. That file has been included to ensure that this repository can be used with Binder (see https://mybinder.org for details).
Please see the following URL for the Binder docker for this tutorial:
https://mybinder.org/v2/gh/adrianbevan/TensorFlow-Tutorial.git/main
AB 8th Oct 2020