Artificial intelligence / Machine learning / Data science / Computer vision / Academic research / Experimental & Archive
Machine Learning Notebook Archive
A source-cataloged archive of 33 independent notebooks exploring machine-learning ideas across image, text, audio and tabular data tasks.
Original code-generated category illustration; not a product screenshot.
Overview
The 32 numbered experiment directories and an additional root notebook combine neural-network fundamentals, image classification, object detection, sentiment analysis and competition modeling. A source-based catalogue corrects misleading historical directory labels. The notebooks are independent educational and research records, each with its own dataset, model and environment requirements.
My role: Learning and experimental work preserved in the repository. Per-notebook contribution or team responsibility is not documented.
Core components
- A source-based subject catalogue covering all 33 notebooks.
- Vision exercises using CNNs and transfer learning.
- LSTM, sentiment-analysis and small translation experiments.
- Speech-emotion, explainability and tabular-data exercises.
Validation and current status
- Notebook JSON and source cells were inspected; no training or API calls were run.
- An incomplete expression and incorrect classifier import were documented in notebook 31.
- The catalogue PR is open; there is no shared automated suite or new benchmark.
- Missing data/model files prevent running every notebook from a clean checkout.
- Privacy review of retained health-research outputs and notebook metadata is incomplete.
- The portfolio publishes no real health outputs, personal records or performance percentages.
Show additional technical details
Problem and objectives
Directory names in notebooks written at different times do not always reflect their actual tasks. Environment, dataset and model requirements also vary. This makes the collection difficult to navigate and obscures the inputs required to reproduce a selected experiment.
Solution approach
A catalogue derived from source cells describes each notebook’s subject, observed libraries and external artifact requirements. Vision, NLP, audio and tabular work are presented as independent experiments. Historical outputs remain preserved without implying successful training, uniform reproducibility or a new comparable performance benchmark.
Architecture flow
- 33 independent Jupyter notebooks.
- Experiment-specific data-loading and preprocessing cells.
- Task-specific classical ML, Keras/PyTorch or vision workflow.
- Notebook-local evaluation, visualization and artifact output.
Technical challenges
- Dependence on different datasets, model artifacts and Colab paths.
- A common requirements list without verified compatibility across every experiment.
- Distinguishing incomplete code and historical outputs from runnable experiments.
Engineering decisions
- Historical directories and source paths are preserved without renaming.
- The catalogue replaces automatic task/model assumptions based on folder names.
- Each experiment requires its own environment and authorized data.
Outcomes
- A multidisciplinary learning archive with clearer subjects and external dependencies.
- Saved charts and metrics are historical artifacts, not newly reproduced results.
The technical account also draws on documentation in the open improvement PR. This does not imply that the PR has been merged or the product released.