Data Science Stuff

Data Science stuff that I find interesting are here. Please ⭐ the repo, if you find it interesting.

Machine Learning

• BatchNorm
• AutoEncoders - Compares PCA vs AutoEncoders. Working example for Dense & Convolutional AutoEncoder
• Variational AutoEncoders - Compares AutoEncoders vs VAEs for generative modelling & their latent spaces. Implements CVAE from tensorflow tutorial & visualizes the latent spaces
• WhiteDistance - Metric to calculate similarity between texts

Model Interpretation

• LIME
• Occlusion
• CAM
• Grad CAM

Neural Arithmetic Logic Unit (NALU)

NLP

• TextGeneration - Copy of Karpathy's work really
• Seq2Seq - Notebook for a word based Encoder-Decoder Seq2Seq network for English to Spanish translation
• Attention - An extension of previous Seq2Seq work which uses attention mechanism for english to spanish translation
• Transformers - Implements individual components of a transformer from scratch
• BertTextClassification - Using BERT for fake news classification. Dataset was obtained from Kaggle here

Optimization

Differential Evolution

Will add more methods as I study them

Cool Stuffs

• stitch_images - Script to create a single image stitched horizontally or vertically from a list of images
• csv_stats.sh - Simple shell script to give summary of all csv files
• code_snipets - Utility Script

Translate App

A simple app which translates english sentence to spanish. It uses an attention based Seq2Seq model. Try and run this app on your local machine -

Streamlit App

Streamlit is an open source framework that let's users create web apps very easily.

1. Install streamlit
2. Clone the repo & navigate to translate_appdirectory
3. Run the app

streamlit run streamlit_app.py

4. Go to localhost:8051

Sanic App

1. Install tensorflow 2.0 and sanic
2. Clone the git repo
3. Navigate to translate_app directory
4. Run server on localhost:5000/translate.

python server.py

5. Evaluate by making curl request at the above url with input_sentence as a parameter. Pass in the desired English sentence as value with it.

curl -X POST -H "Content-Type: application/json"  -d '{"input_sentence":"What did you decide?"}' "localhost:5000/translate"

6. Alternatively, you could run the script make_requests.py and change the sentences in sentences list.

More interesting stuff to follow