Skip to content

Latest commit

 

History

History

classic_ml

README.md

CS-433 Project 2, 2019, Text Classification

File structure

  • cross_validation.py: Module containing implementation of hyperparameter validation and classifier training and scoring
  • feature_extraction.py: Module containing implementations of the procedures required for extracting the different types of features from the processed tweet texts
  • feature_selection.py: Module containing implementations of the feature normalization and selection procedures
  • plotting.py: Script containing code to generate the plots
  • preprocessing.py: Module containing helper functions for performing preprocessing on the tweet text
  • resources.py: Module containing resources reused in multiple scripts, for efficient importing
  • serialization.py: Module containing helper functions for (de)serializing and (de)compressing Python objects from/to local storage
  • training.py: Training script
    • Performing preprocessing and feature extraction on the training set of tweets
    • Performing feature normalization and training a feature selection model using the training feature vectors
    • Performing cross-validation on the classifiers to fit the hyperparameters and find the best model
  • run.py: Main script
    • Performing preprocessing and feature extraction on the testing set of tweets
    • Normalization and feature selection using the pre-fitted transformers
    • Generating the baseline submission using the trained best classifier's predictions
  • files/: Folder with intermediate datasets, serialized transformers and models
  • results/: Folder with final scores, plots and submission file

Reproducibility

Prerequisite

  1. Python 3.7 is installed

  2. That the data is downloaded and extracted like :

├── data
│   ├── train_neg.txt
│   ├── train_pos.txt
│   ├── test_data.txt
  1. That at least these files are present in the files/ folder (as it is now):
├── files
│   ├── best_model.gz
│   ├── test_dataset_reduced.tsv

Notes on dependencies

  • pandas: Data structures and analysis tools. Used for convient representation of tabular data and results.
  • seaborn: Python visualization library based on matplotlib. Used for plotting.
  • nltk: NLP framework. Used for tweet preprocessing and feature extraction tasks.
  • scikit-learn: Machine learning framework. Used for the feature selection, cross-validation, training and evaluation of the standard machine learning classifiers.
  • empath: Tool for analyzing text across lexical categories. Used for feature extraction.
  • _pickle and compress_pickle: Used for Python object serialization and compression to local storage.

Run the code

From the root folder of the project

In order to directly generate the classic ML submission again (it is saved at results/submission.csv):

python3 run.py

In order to perform the feature extraction and selection, cross-validation and classifier training on the train set again:

python3 training.py

In order to perform the feature extraction and selection on the test set again, delete the file test_dataset_reduced.tsv currently present in the files/ folder and run again:

python3 run.py

Authors (team: Definitely not GRUs)

  • Louis Amaudruz
  • Andrej Janchevski
  • Timoté Vaucher