ML-Collections

I spend my free time training and playing around with different neural network architectures. All the below codes use PyTorch as I feel it more flexible.

Perceptron

• From Scratch implementation - Using NumPy functions.
• From Scratch implementation - Using PyTorch functions.

Linear Regression

• Toy Dataset implementaion - Using PyTorch functions.
  • Haven't utilized torch.nn class to create model, i.e hardcoded loss function and layer definition.
• Toy Dataset implementaion - Utilized torch.nn class
  • Used PyTorch's function for loss function, parameter updation and layer definition.

Logistic Regression

• Bank Note Autentication implementation
  • Have used glorot initialisation
  • Remarks - non standardised train data reached higher accuracy quickly, but wasn't steady

MultiLayer Perceptron

• Classification of Flowers using iris Dataset - Implementation.
  • Created model by subclassing nn.Module.
  • Saving and reloading model architecture and parameters.
• XOR Classification - Implementation.
  • Created model with nn.Sequential
    • Easy to create cascading layered models.
  • Created model with nn.Module(reason commented)
    • Helpful for creating complex models which is not strictly Sequential.
  • Created model with nn.Module along with a custom layer.
• Regression model for Fuel Prediction - Implementation.
  • Prepocessed the data from web
  • Trained a Regression model
• Classification of MNIST Digits - Implementation.
  • Used data from torchvision.datasets
• Loss functions and its different inputs - Implementation.
  • Logits and Probabilities as input in,
    • Binary Cross Entropy Loss
    • Cross Entropy Loss

Convolutional Neural Network

• Naive implementation of 1-D & 2-D Convolution in numpy - Implementation
  • Based on the mathematical convolution process, not the Cross Correlation
• Classification of MNIST Digits using CNN - Implementation
  • Accuracy of 99%
• Smile classification using CelebA dataset - Implementation
  • Performed data augmentation pipeline. (Doubt in random retrival of image)
• Eye Glass classification using CelebA dataset - Implementation
  • Played around by tweeking various parameters.
  • Observations are noted.

Transfer Learning

• Check on here.