https://github.com/gnrlleclerc/inf-554-classification-data-challenge

Machine and Deep learnning Kaggle Competition
https://github.com/gnrlleclerc/inf-554-classification-data-challenge

Last synced: about 9 hours ago
JSON representation

Machine and Deep learnning Kaggle Competition

• Host: GitHub
• URL: https://github.com/gnrlleclerc/inf-554-classification-data-challenge
• Owner: GnRlLeclerc
• Created: 2023-12-03T00:20:31.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-07T18:44:49.000Z (about 1 year ago)
• Last Synced: 2024-11-20T14:34:11.529Z (about 2 months ago)
• Language: Jupyter Notebook
• Size: 2.88 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# INF554 Data Challenge

https://www.kaggle.com/competitions/inf554-extractive-summarization-2023/data

## Installation

Using python virtual environment :

```bash

python --version # Check that you have python >=3.11

python -m venv venv

# On MacOS / Linux

source venv/bin/activate

# On Windows

venv\Scripts\activate

# Install dependencies

pip install -r requirements.txt

```

## Usage

### Data

All training data files (`json` utterances and `.txt` graph edges) must be located in the [`data/training/`](./data/training/) directory. The same goes for testing data files in the [`data/test/`](./data/test/) directory.

### Generate utterance text embeddings

We used pretrained embedders from the `sentence_transformers` library.

You can test the performance of our own custom embedder with the following command (we ended up not using it because it would always overfit)

```bash

python src/word_embedding.py

```

In order to save time when embedding utterance, you must first generate the preembedded tensors for each data file. This can be done with the following command :

```bash

python src/loader.py

```

This generates prembedded tensors using 2 models : `all-MiniLM-L6-v2` (small) and `all-mpnet-base-v2` (large).

### Run the classifiers

We experimented on different methods to classify utterances as important or not. The best results were obtained with a simple logistic regression on the embeddings. Our RNN and GNN attempts as well as reimplementations of lab8 and lab9 adapted for this task can be found in the [`rnn.py`](./src/rnn.py), [`graph_model.py`](./src/graph_model.py), [`lab_lstm.py`](./src/lab8_lstm.py) and [`lab9_gnn.py`](./src/lab9_gcn.py).

You can run the logistic regression classifier with the following command :

```bash

python src/majority.py

```

## File structure

```bash

├── .vscode         # VSCode configuration

│

├── data            # Datasets

│   ├── test        # Data for testing

│   └── training    # Data for training

│

├── labels          # Computed labels + true training labels

│   ├── test        # Labels for testing

│   └── training    # Labels for training

│

├── notebooks       # Jupyter notebooks (experiments)

│

└── src             # Python script dependencies

```

## Usage

Run the baseline text-only (no graph) model with the following command.

It will test it against training and validation data, and write to labels/train the estimated test labels.

```bash

python src/baseline.py

```

Use the following command to display help about the function that generates a submission file.

```bash

python src/make_submission.py -h

```