Deep Learning Workshop

This repository is for KKStream Deep Learning Workshop.

During the workshop, you would use, at least, two datasets: Fashion MNIST, and KKTV Data Game 17.11. The famous open-source libraryies, TensorFlow and Keras, would get used, but in some step-by-step ways such that you don't need have experience with them. Furthermore, some fundamental mathematics would get practiced through codes, including Linear Algebra (matrix operation), Calculus (differential) and even Linear Regression. Finally, the mythical Deep Learning would be guided with some hints about parameters you can tune.

Oh, do we mention that the whole workshop requires you do some Python coding?!

Expectations

• learn the basic concepts of deep learning
• learn the general structure of neural networks
• learn the general structure of convolution neural networks
• achieve auc 0.85 on KKTV Data Game 17.11

Installation

There are, at least, two ways confirmed to prepare and install your environment for this workshop project. The 1st one is through the classical Virtualenv and Pip. The 2nd is to leverage the new proposal of Python package management, Pipenv.

Oh, forgot to mention that the workshop is tested on Python 2.7. We haven't examed it on Python 3, but you can try and let us know :)

1. Virtualenv & Pip

• virtualenv (if you don't know what virtualenv is, you definitely need it.)
• clone this repo
• cd datateam_workshop_dl
• virtualenv venv (-p /path/to/python2 if it's python3)
• source venv/bin/activate
• pip install jupyter
• pip install tensorflow or pip install tensorflow-gpu if you have gpu
• pip install keras
• pip install matplotlib
• pip install scipy
• pip install sklearn
• pip install git+https://www.github.com/keras-team/keras-contrib.git
• download datasets:
  • ./datasets/v0_eigens.npz
  • ./datasets/sample.png
  • ./datasets/inception.png
• jupyter notebook
• explore the notebooks in your browser.

2. Pipenv

In this installation through pipenv, package tensorflow is installed, instead of tensorflow-gpu. You can manually update the Pipfile before running through the following steps then.

• install pipenv through the official doc.
• clone this repo
• cd datateam_workshop_dl
• pipenv install (that's it!)
• download datasets:
  • ./datasets/v0_eigens.npz
  • ./datasets/sample.png
  • ./datasets/inception.png
• pipenv run jupyter notebook
• explore the notebooks in your browser.

Source Code

• 00_requirements.ipynb : run all cells to meet all requirements.
• 01_tensors.ipynb : tensor basics.
• 02_handicrafts.ipynb : a handcrafted shallow neural network.
• 03_dense.ipynb : basic neural networks on keras.
• 04_convolution.ipynb : what convolution neural network is.
• 05_convnet.ipynb : basic convolutional neural networks on keras.

Before the Workshop

• clone the repo and install everything.
• make sure all cells in 00_requirements.ipynb work fine.
• join the workshop competition on kaggle.
• try to understand 01_tensors.ipynb before the workshop (optional).