Sentiment Analysis Pipeline

This project contains a complete pipeline of experiments using DVC, Docker and wandb . The instructions below will allow you to build the environment and reproduce the results.

---

🐳 Run via Docker

1. Build a Docker image

``bash make build

### 2. Create an .env file with the API key for Weights & Biases
WANDB_API_KEY=your_key_api
### 3. Download the data and deploy it to the
🗂️ rt-polarity
- Download the data from: http://www.cs.cornell.edu/people/pabo/movie-review-data/rt-polaritydata.tar.gz
- Extract the contents to:
```bash
data/raw_data/rt-polarity/

• So that the files inside are:

rt-polarity.neg
rt-polarity.pos

🗂️ sephora

• Download the data from: https://www.kaggle.com/datasets/nadyinky/sephora-products-and-skincare-reviews
• Extract all .csv files to:

data/raw_data/sephora

• By doing the project on custom dataset, rename the "rating" column from Reviews data to "LABEL-rating".

4. Start a Docker container with access to the environment

make run_docker

5. Inside the container, execute the pipeline

dvc repro

📁 Configuration structure

folder structure should be mapped in case of deficiencies

• params.yaml - main parameter file, shared among all configurations. Contains model settings, features and general hyperparameters.
• configs/sephora.yaml, configs/rt-polarity.yaml - dataset-specific configurations (paths, columns, pipeline parameters).
• data/raw_data/ - directory where you should manually drop the raw input data into:
• sephora/ with CSV files from Kaggle
• rt-polarity/ with files rt-polarity.neg and rt-polarity.pos
• data/processed_data/ - data loaded from raw_data, processed into pkl, ready for analysis and preprocessing.
• data/preprocessed_data/ - data after preprocessing.
• data/train_test_split/ - files containing splits for training and test collection.
• data/models/ - saved models for each configuration along with the best parameters in .json files.
• data/metrics/ - evaluation metrics saved for each experiment.
• data/notebooks/ - data generated during exploration in notebooks.
• notebooks/ - data mining, EDA analysis, hyperparameter search and sharp analysis.
• excercises/ - directory for exercises/tests.
• results/ - final tables of results comparing models/datasets.
• reports/ - graphs, tables or other final artifacts for the report.
• scripts/ - auxiliary scripts (e.g. for downloading, cleaning data).
• src/ - main project code (transformers, tool functions, pipeline).

📝 Notes

Pipeline automatically performs all processing and saves the results without the need for additional commands. It is required to have an account in Weights & Biases and set your own API key in .env.

✅ End result

After dvc repro:

• Models will be saved in models/{dataset_name}
• Results will be saved in results/test_results.md
• Detailed metrics in data/metrics/*