Projects completed as part of the Udacity Deep Learning Nanodegree program.
Built a neural network which predicts daily bikesharing patterns. Predictions made are done using past data on the number of riders for each hour of each day from January 1 2011 to December 31 2012. The network built is a two layer network with a hidden layer and output layer. A sigmoid function is used for activations. The training goes through 1350 iterations, uses a learning rate of 0.55, has 3 hidden_nodes and 1 output node.
Below shows a graph comparing the real data with the patterns predicted through the neural network:
Wrote an algorithm for Landmark Classification. The project is composed of 4 parts. The algorithm was primarly written from scratch.
- Download Datasets and Install Python Modules
- Create a CNN to Classify Landmarks (from Scratch)
- Create a CNN to Classify Landmarks (using Transfer Learning)
- Write Your Landmark Prediction Algorithm
Makes use of LSTMS and TV Scripts from the Seinfeld dataset to generate new TV scripts. The model was ultimately able to generate the scripts with a loss of 3.43. Iterating through more epochs is likely to improve the accuracy of the model in generating realistic and feasible scripts.
Makes use of a Deep Convolutional Generative Adversarial Network (DCGAN) on a training set of celebrity faces to generate fake faces. The project consists of data loading and pre-processing, creating the generator and discriminator classes, as well as training the network. After 50 epochs, fake faces are generated at the end of the file.
Train and deploy your own PyTorch sentiment analysis model. Deployment gives you the ability to use a trained model to analyze new, user input. Build a model, deploy it, and create a gateway for accessing it from a website.
Udacity, 2021