machine learning and deep learning
Learn machine-learning and deep-learning.
Sources
- Code on Github: lazyprogrammer/machine_learning_examples
- Deep-learning-python-part1: data-science-deep-learning-in-python
- Facial-expression-recognition:
- Code on Github: lazyprogrammer/facial-expression-recognition
- Dataset: Kaggle facial expression
- Deep-learning-python-part2: theano-tensorflow
- Dataset: Kaggle digit recognizer
- 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/
- 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.
- For
backprop.py
andann_train.py
Learn how to rewrite softmax
, forward
, predict
, classification_rate
, and cross_entropy
into compact matrix form.
- For
logistic_softmax_train.py
andann_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.