Implementation of deep neural networks from scratch for multiclass classification with mean squared error and cross entropy loss, tanh, relu and sigmoid activations, gradient descent, momentum, nag and adam optimizers with l2 regularization.
- Numpy
- Pickle
- Sklearn
Download the directory and use bash run.sh in terminal. Change placeholders in run.sh for train, validation and test data. Weights will be stored in save_dir and logs will be stored in expt_dir along with predictions.
Implementation of CNN for multiclass classification in Tiny Imagenet dataset. Also includes code for fooling network, guided backpropogation and plotting weight of layer 1 filter.
- Numpy
- Pickle
- Sklearn
- Tensorflow
- scipy
- Matplotlib
- Pandas
Download the directory and use bash run.sh in terminal. Change placeholders in run.sh for train, validation and test data.
Implementation of RNN for transliteration from english to Hindi. Also includes code for bidirectional lstm and attention.
- Numpy
- Pickle
- Tensorflow
- Matplotlib
- Scipy
Download the directory and use bash run.sh in terminal. Change placeholders in run.sh for train, validation and test data.