cenyunuo/ee5907_2020fall_ca2

EE5907_2020Fall_CA2

Introduction

This is the readme file for desrbibing the codes for EE5907 CA2.

In this course project, we eavluate several algorithms: PCA/LDA/GMM/SVM/CNN, on CMU PIE dataset.

Here is the content of this project. Please do arrange your documents in such way:

├── Readme.md                   
├── PIE
│   ├── 1
│   ├── 2
│   ├── ...
│   └── 26
├── main.py
└── common.py    

Preliminaries

Running this code requires:

1. numpy
2. tensorflow
3. matplotlib
4. sci-kit learn
5. common.py (Some common functions required in main.py, but seperate to another document to make code looks neat).
6. Make sure you include the dataset in the folder PIE in the same path. And make sure the total class of the dataset should be 26.

Usage

This code can support 6 algorithm for evaluate the PIE dataset. You can run the codes like this:

python main.py -a ... -m ... -d ... -p ...

Since we have totally 5 algorithms here, and they have different methods and parameters, so we should specify them when we run the codes.

• -a: Algorithm, where you should specify whether is PCA/LDA/GMM/SVM/CNN;

• -m: Method, where you should specity whether is what method you want to evaluate on the algorithm;

• -d: Dimension, where you should specify how many reduced dimensions you want to retain (-d is not required for CNN);

• -p: The parameters you want to apply (-p is not required for PCA, LDA and GMM).

• methods of PCA:

1. vis: Visualize the PCA result in 2 or 3 dimensions;
2. face: Reconstruct faces from reduced PCs;
3. classify: Classify with KNN classifiers with reduced PCs.

• methods of LDA:

1. vis: Visualize the LDA result in 2 or 3 dimensions;
2. classify: Classify with KNN classifiers with reduced PCs.

• methods of GMM:

1. clustering: Visualize the clustering result of GMM with reduced PCs;

• method of SVM:

1. classify: Classify with SVM classifiers with reduced PCs. And you should specify -p with penalty you want to apply.

• method of CNN:

1. classify: Classify with CNN model. And you should specify -p with epochs you want to train the model.

Example

For exmaple, you want to visulize the PCA result in 2 dimensions:

python main.py -a PCA -m vis -d 2

Or, you want to train a SVM model with 40 reduced PCs and penalty of 0.1:

python main.py -a SVM -m classify -d 40 -p 0.1

Or, you want to train the CNN with 20 epochs:

python main.py -a CNN -m classify -p 20