gaopinghuang0/machine-learning

Stuff on machine learning and deep learning

machine learning and deep learning

Learn machine-learning and deep-learning.

Sources

1. Code on Github: lazyprogrammer/machine_learning_examples
2. Deep-learning-python-part1: data-science-deep-learning-in-python
3. Facial-expression-recognition:
   1. Code on Github: lazyprogrammer/facial-expression-recognition
   2. Dataset: Kaggle facial expression
4. Deep-learning-python-part2: theano-tensorflow
   1. Dataset: Kaggle digit recognizer
5. Deep-learning-python-part5: Recurrent neural networks

Run

# if virtualenv or python3 is installed
python3 -m venv env   # where env is a conventional folder name to store all the scripts and dependencies
# or 
virtualenv env

# activate
source env/bin/activate   # in my .bashrc, alias to `workon`

# install using pip3, which comes with virtualenv
pip3 install -r requirements.txt   # which comes from  pip freeze > requirements.txt

# exec any python script
python3 script.py

Notes:

Under deep-learning-python-part1/

1. For backprop.py

Learn how to rewrite slow for-loop into fast matrix form. Also, we can learn how to update W2, b2, W1, b1 using backpropagation.

2. For backprop.py and ann_train.py

Learn how to rewrite softmax, forward, predict, classification_rate, and cross_entropy into compact matrix form.

3. For logistic_softmax_train.py and ann_train.py

logistic_softmax only involves W and b. So the forward mode is softmax(X.dot(W) + b), and the gradient decent only updates W and b.

ANN (one-hidden-layer) forward involves W1, b1, W2, b2, while backpropagation updates all of them.