udacity-deep-learning

Projects and exercises from the Deep Learning course at Udacity

Overview

This repository serves as a container for archiving all the work completed for the Deep Learning course at Udacity. This is meant to keep things tidy and organized, with all the related code in one place. Originally, each project had its own repository, but it led to too much clutter, as Github doesn’t currently offer any means of grouping one’s repos.

Contents

Projects

1. Predicting Bike-Sharing Patterns
   • Manually building a simple MLP from scratch.
   • Training the network for a regression task.
   • Documenting the hyperparameter tuning process.
2. Dog-Breed Classifier
   • Building a CNN for image classification using transfer-learning and PyTorch.
   • Image data pre-processing.
   • Discriminating between human and dog images, as well as classifying dog breeds.
3. Generate TV Scripts
   • Pre-processing text data in the form of TV scripts.
   • Building and training a LSTM RNN.
   • Generating novel TV scripts (mainly dialogues).
4. Generate Faces
   • Building a Generator and Discriminator modules to form a GAN network.
   • Preparing a custom training process.
   • Generating low-res images of human faces.
5. Deploying a Sentiment Analysis Model
   • Creating and configuring an AWS account.
   • Building a movie review sentiment analysis model using SageMaker.
   • Deploying the model using endpoints, Lambda and API Gateway.

Exercises

• Various smaller projects, often involving building and training different NN architectures and learning key ML concepts.
• Mostly done in Jupter and Pytorch.
• Some use Tensorflow + Keras as well.