This repository holds some deep learning projects I have worked on in class.
I built a neural network from scratch using only numpy and trained it on the dataset from the famous Titanic challenge on Kaggle. Here is the submission:
With the help of diegoalejogm's implementation of a Vanilla GAN generating hand-written digits (https://github.com/diegoalejogm/gans/blob/master/1.%20Vanilla%20GAN%20PyTorch.ipynb), I used a dataset of small flag icons from flagpedia (http://flagpedia.net/download) to generate small black and white flag icons.
Examples of the training data:
Examples of the resulting images (after 8000 epochs):
This Pytorch LSTM and GRU perform multi-class categorization on online comments from a kaggle challenge (https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge). The comments are categorized in the following six categories:
- toxic
- severe toxic
- obscene
- threat
- identity hate
These models were built after the end of the challenge, so the results are based on the actual test set data, which was made public after the deadline.
In this notebook, I am applying the pretrained model from Jeremy Howard & Sebastian Ruder's paper "Universal Language Model Fine-tuning for Text Classification" (2018, see https://arxiv.org/abs/1801.06146) to the twitter airline sentiment dataset from Kaggle (https://www.kaggle.com/crowdflower/twitter-airline-sentiment/downloads/twitter-airline-sentiment.zip/2). In my implementation I am borrowing from fastai's tutorial on ULMFiT (https://docs.fast.ai/text.html).