CS771-Mini-Project

This respositry consists of the submission for the First Mini Project for the course CS771, Fall 2024, completed under the instruction of Prof. Piyush Rai, Department of CSE, IIT Kanpur

Team

Name	Roll Number
Anushka Singh	220188
Arush Upadhyaya	220213
Aujasvit Datta	220254
Pahal Dhruvin Patel	220742
Pranav Agrawal	220791

Contents

• 17.py : main file to generate and save predictions
• utils.py : utility functions used in 17.py
• pred_emoticon.txt : predictions for the emoticons dataset
• pred_deepfeat.txt : predictions for the deep features dataset
• pred_text_seq.txt : predictions for the text sequences dataset
• emoticons/ : jupyter notebooks containing experiments and EDA for emoticons dataset
• features/ : jupyter notebooks containing experiments and EDA for features dataset
• text_seq/ : jupyter notebooks containing experiments and EDA for text sequences dataset
• combined/ : jupter notebooks containing experiments and EDA for all datasets combined
• common/ : helper functions used in experiments

Instructions

• Install the dependencies

pip install -r requirements.txt

• Download the dataset, make sure the datasets/ directory is present in the root

• Run 17.py to generate the prediction files →

python 17.py

Approaches

Dataset-1

• Preprocessing :

  1. Removed dummy emojis, that are occuring in all the input emoji strings
  2. Columnarised the emoji strings into one column per character
  3. One hot encoded the categorical columns

• Model : Logistic Regression

• Best Parametres

  Parameter	Value
  C	10
  penalty	L1
  Solver	Liblinear

• Achieved Accuracy on Validation Set : 97.13%

Dataset-2

• Preprocessing : None

• Model : Logistic Regression

• Best Parametres

  Parameter	Value
  C	10.0
  fit_intercept	True
  penalty	l2
  solver	lbfgs

• Achieved Accuracy on Validation Set : 98.77%

Dataset-3

• Preprocessing

  1. Removed substrings occuring in all the input strings
  2. Converted the input strings into n-gram respresentation, with $n_range = (3, 5)$

• Model : Logistic Regression

• Best Parametres

  Parameter	Value
  colsample_bytree	1.0
  eval_metric	logloss
  gamma	0.2
  learning_rate	0.1
  max_depth	7
  min_child_weight	3
  n_estimators	500
  subsample	1.0

• Achieved Accuracy on Validation Set : 93.05%

Task 2

• Model : Logistic Regression

• Best Parametres :

  Parameter	Value
  C	10.0
  fit_intercept	True
  penalty	l2
  solver	lbfgs

• Achieved Accuracy on Validation Set : 98.77%

We used the seed 42 for all the probabilistic models that we attempted to run.