Collection of scripts I use in Classroom
python -m venv .venv
source ./.venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python -m ipykernel install --user --name deeplearning- Open this folder in VS Code, install the Jupyter and Python extensions.
- Open
*.ipynb. - Select the
deeplearningkernel (or.venv) via the kernel picker (top-right). - Run cells with
Shift+Enter, or Run All.
| SN | Topic | Script |
|---|---|---|
| 1 | Introduction to Pytorch (with toy example) | GO |
| 2 | Dataset splits, Overfit/Underfit and Dropout/L1/L2 regularization | GO |
| 3 | Limitation of Gradient Descent | GO |
| 4 | Adaptive Learning Rates: AdaGrad, RMSProp, Adam | GO |
| 5 | Why do we need CNNs? DNN limitations vs. Convolutions (FashionMNIST) | GO |
| 6 | AlexNet: basic idea, architecture, building in PyTorch, and training on a small subset (CIFAR-10) | GO |
| 7 | Why do we need Batch Normalization? Training dynamics, learning-rate sensitivity, and internal covariate shift, for both an MLP and a CNN (FashionMNIST) | GO |
| 8 | Representation Learning: a simple autoencoder, reconstruction, visualizing the learned code, and transferring it to classification with very few labels (MNIST) | GO |
| 9 | Word2Vec as an autoencoder: skip-gram encoder-decoder, nearest neighbours and a word map, and using the word vectors for document classification with few labels (20 Newsgroups) | GO |
| 10 | Embedding-based retrieval: a frozen EmbeddingGemma encoder, sentence embeddings in a Python dict, cosine nearest-neighbour search, keyword (TF-IDF) comparison and cross-lingual Nepali queries (news corpus) | GO |
| 11 | Why do we need sequence models? Truncation, padding and averaging (bag of words / frames) vs. a pure recurrent model (GRU) on variable-length text and speech | GO |
| 12 | Neural N-Gram language model (Bengio et al., 2003): count-based trigram vs. neural trigram, unseen contexts, next-word prediction, text generation, learned embeddings and the fixed-window limitation (toy corpus) | GO |
The Encoder/Decoder model related notebooks are in Rep-Learning directory.
Note: This list will gradually grow. Hopefully in 2-3 months :-)