https://github.com/atalv/experiments
Experiments done for hands-on learning with dummy data
https://github.com/atalv/experiments
Last synced: about 22 hours ago
JSON representation
Experiments done for hands-on learning with dummy data
- Host: GitHub
- URL: https://github.com/atalv/experiments
- Owner: atalv
- License: gpl-3.0
- Created: 2023-06-21T10:07:11.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2024-01-18T21:08:29.000Z (over 2 years ago)
- Last Synced: 2025-10-29T14:58:07.646Z (8 months ago)
- Language: Jupyter Notebook
- Size: 5.64 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Citation: CITATION.cff
Awesome Lists containing this project
README
# Overview
This repo is to store the experiments done for hands-on learning with *dummy data*. It is never too late to learn and *showcase*!
- Each sub-directory in the root is named as the main topic of the experiment.
- All the contents are created solely by me with guidance from official resources and academic experts.
## Citation
If you use any of this work then please add a referrence to this repository ['Experiments by Vivek Atal'](https://github.com/atalv/experiments/) as a fair usage policy.
## Some highlighs
- **GraphNetwork**:
- Predicted whether a user of LastFM would follow another user and serve as a recommendation.
- Implemented multiple node embedding approaches for link prediction - *[Graph Factorization](https://doi.org/10.1145/2488388.2488393), [DeepWalk](https://arxiv.org/pdf/1403.6652.pdf), [Node2Vec](https://arxiv.org/pdf/1607.00653.pdf), [Adamic-Adar index](http://www.cs.cornell.edu/home/kleinber/link-pred.pdf)* - and compared their performance for link prediction task.
- Referred to excellent materials by Stanford CS224W course on [Machine Learning with Graphs](http://web.stanford.edu/class/cs224w/).
- **MachineLearning**:
- Predicted NYC taxi trip duration.
- Implemented typical machine learning models from *[scikit-learn](https://scikit-learn.org/stable/)* ([GammaRegressor](https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.GammaRegressor.html), [RandomForestRegressor](https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestRegressor.html), [HistGradientBoostingRegressor](https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.HistGradientBoostingRegressor.html)) with intelligently derived features, viz., traffic information in an area at a given time window based on average active number of trips originating or ending.
- **ReinforcementLearning**:
- Simulated multiple UCB (Upper Confidence Bound) policies for MAB ([Multi Armed Bandit](https://en.wikipedia.org/wiki/Multi-armed_bandit)) problems and MDP ([Markov Decision Process](https://en.wikipedia.org/wiki/Markov_decision_process)) and compared their performance.
- Learned to do simulation of multiple states Markov Chain and calculate average reward, expected present value, estimate steady state probabilities, etc.
- Most of the research papers referred for simulation exercises are authored by [Dr. Michael Katehakis](http://en.wikipedia.org/wiki/Michael_N._Katehakis).
- **TimeSeries**:
- Forecasted 2 weeks ahead grocery store sales of 33 product groups across 54 stores, approx. 1.8K time series.
- Engineered multiple sensible features, viz., cross-store, cross-product elements, algorithmically short-listed important events for a given store-product, etc.
- Some Seasonal ARIMA models were built manually, and then scaled it using ARIMA where seasonal components were extracted beforehand for faster execution.
- Experimented with [DeepAR on AWS Sagemaker](https://docs.aws.amazon.com/sagemaker/latest/dg/deepar.html) to build a single global model instead of 1.8K ARIMA models.