bholloway/ml_class

DeepSchool.io

Deep Learning Learning

Goals

1. Make Deep Learning easier (minimal code).
2. Minimise required mathematics.
3. Make it practical (runs on laptops).
4. Open Source Deep Learning Learning.

Installation

1. Install Docker https://www.docker.com/
2. Use the following commands to run from docker¹.

git clone git@github.com:sachinruk/deepschool.io.git
cd deepschool.io
docker-compose up --build

3. Now go to localhost:8888 on your browser to start using the jupyter notebooks.

Contents

The lessons will cover the fundamentals of deep learning.

0. Lesson 0: Introduction to regression.
1. Lesson 1: Penalising weights to fit better (scikit learn intro)

Mathematics (optional)

2. Lesson 2: Gradient Descent. Using basic optimisation methods.
3. Lesson 3: Tensorflow intro: zero layer hidden networks (i.e. normal regression).
4. Lesson 4: Tensorflow hidden layer introduction.

Deep Learning

5. Lesson 5: Using Keras to simplify multi layer neural nets.
6. Lesson 6: Embeddings to deal with categorical data. (Keras)
7. Lesson 7: Word2Vec. Embeddings to visualise words. (Tensorflow)
8. Lesson 8: Application - Bike Sharing predictions
9. Lesson 9: Choosing Number of Layers and more
10. Lesson 10: XGBoost - A quick detour from Deep Learning
11. Lesson 11: Convolutional Neural Nets (MNIST dataset)
12. Lesson 12: CNNs and BatchNormalisation (CIFAR10 dataset)
13. Lesson 13: Transfer Learning (Dogs vs Cats dataset)

Notes

1: Refer to this Dockerfile and this for information on how the docker image was built.