Machine_Learning

Using the SVHN database. We have logistic regression(+Lasso, +Ridge), SVM(+kernel), GMM, LDA, GAN, CNN, and so on.

Installation

git clone https://github.com/shiyegao/Machine_Learning.git
cd Machine_Learning
pip install -r requirements.txt

Usage

Examples

For example, you can run logistic regression (lr) with lasso loss using full dataset as follows.

python main.py --method lr --mode lasso --data vec --dataset full

Besides, you can run GAN using full dataset as follows.

python main.py --method gan --data img --dataset full

Parameters

There are some parameters which decides the dataset to choose from.

• --data, choices=['vec', 'img'], whether to use HOG features, or RGB features

• --dataset, choices=['full', 'part'], whether to use full dataset, or the part

There are some parameters which decides the model to choose from.

• --method, choices=['svm','lr', 'lda', 'gmm', 'cnn', 'gan']
• --mode, choices=['normal','lasso', 'ridge']
• --kernel, choices=['linear', 'poly', 'rbf', 'sigmoid', 'precomputed']

There are some parameters which decides the hyper-parameters to choose from.

• --lr, learning rate
• --Lambda, used for Lasso and Ridge"
• --epoch, epoch for training"
• --batch-size, choose -1 for the whole dataset"

More details can be seen in utils.py.