Week 3 - Intro Kaggle competition - EDA and baseline models with PyTorch

Learning and testing - a.k.a. don't do Bullshit Machine Learning

Course session

https://drive.google.com/file/d/18VsrKSqNFaWeWsL24ULFrNnwpghpdwJZ/view?usp=sharing

Walk-through

Hyperparameter experiment

The following notebook will show how to set up a hyperparameter experiment in plain PyTorch. More importantly it give you the results and enables you to analyze and play around

Google Colabcolab.research.google.com

Kaggle

• Introduction

• Titanic

Solutions exercise MLP

Presentation from the participants of the MLP from Coursera

To-do

😊

Watch the videos on the next page

Go through the following notebooks and complete the second one (assignment notebook):

Google Colabcolab.research.google.com

The first notebook

Google Colabcolab.research.google.com

The assignment notebook

The next task is to analyze the results of the hyperparameter experiment and create a small presentation on your findings(e.g. batch size of 16 with lr=0.2 seems to equal batch size of 1 with lr=0.01). Here is the notebook again:

Google Colabcolab.research.google.com

😊😊

Run your own hyperparameter experiment

😊😊😊

Do your own EDA on the Titanic Dataset and/or look at other EDA notebooks from competitors. Make a final presentable EDA notebook.

Familiarize yourself with this PyTorch Tutorials:

Tutorial 5: Inception, ResNet and DenseNet — UvA DL Notebooks v1.2 documentationuvadlc-notebooks.readthedocs.io