| Applied mathematics, Data Science

Artificial Intelligence Frameworks

This course follows the Machine Learning and the High Dimensional & Deep Learning courses. In theses courses, you have acquired knowledge in machine and deep learning algoritms and their application on various type of data. This knowledge is primordial to become a DataScientist.

This course has three main objectives. You will

• learn how to apply efficiently these algorithms using

* Cloud computing with Google Cloud, * Container with Docker,

• discover new field of artificial intelligence applied on (real) datasets that require specific algorithms:

* Text. * Algorithms: Text processing, Vectorizer, Words Embedding, RNN Libraries : Nltk, Scikit-Learn, Gensim* * Video Game * Algorithms: Reinforcement learning, (Policy Gradient algorithm, Q-Learning, Deep Q-learning) Libraries : AI Gym, Tensorflow. * Movies Notations * Algorithms: Recommendation system, (User/User and Item/Item filters, NMF, Neural recomendation system) Libraries : Surprise, Tensorflow.

• how to efficiently share reproducible code.

* Build a Github repository.

NB: Some contents from previous years are still available on the repository (like Spark) but are not covered during theses courses anymore.

Knowledge requirements

• R Tutorial
• Python Tutorial
• Elementary statistic tools
• Data Exploration and Clustering.
• Machine Learning
• High Dimensional & Deep Learning

Schedule

• Lectures : 10 hours
• Practical Works : 30 hours.

The course is divided in 5 topics (of various lentgh) over 5 days.

Course introduction + Github Reminder: Slides/Video

• Session 1 - 02-11-20

- Text: Text Cleaning + Text Vectorization * Slides / TP / Video - Text: Words Embedding. * Slides / TP / Video

• Session 2 - 16-11-20

- Text: Recurrent Network * Slides / TP / Video - Development for Data Scientist: Python environment + Github Repo + Python Script. * Slides / TP / Video

• Session 3 - 30-11-20

- Development for Data Scientist: Introduction to Google Cloud Computing. * Slides / TP / Video - Development for Data Scientist: Docker * Slides / TP / Video

• Session 4 - 07-12-20

* Introduction to deep Reinforcement learning: Deep Q-learning - Slides / Video - Q Learning reminder: TP - Deep Q Learning on cartpole: TP - Deep Q Learning on Gridworld: TP

• Session 5 14-12-20

- Introduction to deep Reinforcement learning: PG Gradient * Slides / TP / Video - Recommendation System. * Slides / TP1 & TP2 / Video

• Session 6 04-01-20

- Free time on project.

Evaluation

The evaluation is associated to the DEFI-IA

Objective

You will be evaluated on your capacity of acting like a Data Scientist, i.e.

• Handle a new dataset and explore it.
• Find a solution to address the defi's problem with a high score (above baseline).
• Explain the choosen algorithm.
• Write a complete pipeline to easily reproduce the results.
• Justify the choice of the algorithms and the environment (CPU/GPU, Cloud etc..).
• Share it and make your results easily reproducible (Git - docker, conda environment.).

Notations

• Project - (60%): a Git repository.

* The git should contain a clear markdown Readme, which describes (33%) * Which result you achieved? In which computation time? On which engine? * What do I have to install to be able to reproduce the code? * Which command do I have to run to reproduce the results? * The code has to be easily reproducible. (33%) * Packages required has to be well described. (a requirements.txt files is the best) * Conda command or docker command can be furnish * The code should be clear and easily readable. (33%) * Final results can be run in a script and not a notebook. * Only final code can be found in this script. * Deadline : January 29 2021.

• Rapport - (40%) 10 pages maximum:

* Quality of the presentation. 25% * In-Deep explanation of the chosen algorithm. 25% * Choice of the tools-infrastructure used. 25% * Results you obtained. 25% * Date : January 29, 2021.

Other details

* Group of 4 to 5 people (DEFI IA's team). ## Technical requirements. All the libraries required for these modules are listed in the requirements.txt (IN CONSTRUCTION/ ONLY SESSION 1 IS OK) To build a functional environment in pandas execute the following lines:

conda create -n AIF python=3.8 conda activate AIF pip install -r requirements.txt  jupyter labextension install jupyterlab-plotly@4.12.0