Description
1 Submission Guidelines
• Please note that assignment 3 should be submitted via canvas.
• You need to zip the following two files together:
– assignment3 studentid report.pdf/.docx: Please put your report in this file. (Attachments should be original .pdf or .docx, NOT compressed)
– assignment3 studentid code.zip: The zip file contains all your source codes for this project.
• All attachments, including report and code, should be named in the format of: student studentid.zip.
• Submissions not following the rules above are NOT accepted.
• Your grade will be based on the correctness, efficiency and clarity.
• The email for Q&A: [email protected].
2 Objective
The objective of this project is twofold:
• To acquire a better understanding of neural networks by using a publicdomain software package called PyTorch.
3 Major Tasks
This assignment consists of the following tasks:
• To install and learn to use PyTorch.
• To train multilayer perception (MLP) neural network models using all data sets provided, including five binary data sets and one multi-class data set.
• To write up a report.
These tasks will be illustrated in the following sections.
3.1 PyTorch Installation
Detailed guides on installing PyTorch for different computing platforms can be found at the website https://pytorch.org/get-started/locally/. Installing PyTorch using Anaconda or PIP is the recommended way since it is simple and straightforward.
4 Data Sets
4.1 Five Binary Classification Data Sets
You will use five binary classification data sets which are available in the ZIP file (datasets.zip). The table 1 shows the number of features, number of training examples, and number of test examples for each data set.
When you load each .npz data file, you will find four NumPy arrays train X, train Y, test X and test Y. Each row of X stores the features of one example and the corresponding row of Y stores its class label (0 or 1). As is always the case, the class label files for the test sets should not be used for classifier training but only for measuring the classification accuracy on the test data.
Table 1: Summary of Binary Classfication Data Sets
Data Set #features # train # test
Breast cancer 10 547 136
Diabetes 8 615 153
Digit 64 800 200
Iris 4 120 30
Wine 13 142 36
4.2 Multi-class Data Sets
The dataset consists of a training set of 10,000 examples and a test set of 1,000 examples. Each example is a 28×28 grayscale image, associated with a label from 10 classes. Please load the data set with scipy such as from scipy.io import loadmat.
5 Train and Test the Nerual Networks Models
5.1 Neural Networks Models for Binary Classifcation Data Sets
For 5 binary classifcation data sets D “ tD1,D2,¨¨¨ ,D5u, you are required to construct a set of single hidden layer neural network models. The number of hidden units H should be determined using cross validation. The generalization performance of the model is estimated for each candidate value of H P t1,2,¨¨¨ ,10u. This is done by randomly sampling 80% of the training instances to train a classifier and then testing it on the remaining 20%. Hence, given 5 datasets, you are required to perform 5 such random data splits in order to find the most suitable H˚ “ t u for 5 data sets, respectively.
Subsequently, given any binary data set Di, you are requied to train a neural network classifier with Hi˚ hidden units in a single layer, which is trained from scratch using all the training instances available.
5.2 Neural Networks Models for Multi-class Data Sets
For the multi-class data set, you are required to construct a set of two hidden layers neural network models. The number of hidden units for first layer L1 is choosen from t50,75,100u, while the number of hidden units for second layer L2 is choosen from t10,15,20u. Like the single layer NNMs above, the best combination of hidden numbers for each layer is decided by cross validation (randomly sampling 80% of the training instances to train a classifier and then testing it on the remaining 20%).
5.3 Report Writting
6 Tips
6.1 Programming Tips
As is always the case, good programming practices should be applied when coding your program. Below are some common ones but they are by no means complete:
• Using functions to structure program clearly.
• Using meaningful variable and function names to improve readability.
• Using indentation
• Using consistent styles
• Including concise but informative comments
7 Grading Schema
The assignment grading schema has been listed as:
• Coding: (60 marks) Essential comment will be helpful for your grading.
– Build neural network models and adopt the gradient descent optimization algorithm (20 marks).
– Tune the parameters using cross validation techniques (20 marks).
– Compute the cross entropy loss and the accuracy of the neural network model on both the training and test sets (20 marks).
• Project Report: (40 marks)
– Present the experiment settings of the neural network models such as optimizer and learning rate (10 marks).
– Report the parameter tuning result of the neural network model using cross validation (20 marks).
– Report the accuracy of the selected neural network models on the test set (10 marks).
