Pawan300 / NLP

This repository contains the programs related to NLP.

• This contain some research paper implementation or some transformers extension of hugging face in Text similarity.

• I have tried some approaches on the simple dataset which trying to classify the types of text into spam or ham.
  So I have tried mulitple strategy to come up for the embeddings:
  

  • TF_IDF
    
  • Word2Vec
    
  • Doc2Vec
    

  Then I tried random forest and RNN structure with LSTM.
  

  Scores I get is:
  

  Model	Precision	Recall	Accuracy
  TF_IDF + RF	0.99	0.78	0.97
  Word2Vec + RF	0.46	0.24	0.87
  Doc2Vec + RF	0.81	0.35	0.91
  RNN + text_to_sequence	0.99	0.96	0.99

  I also tried to catch some hyperparameter using different methods and libraries :
  

  Model	Time (in min)	Accuracy
  Random forest (RF)	2.4	0.97
  Grid Search CV	25.6	0.97
  Pipeline	10.9	0.95
  Skopt	19.3	0.97
  Hyperopt	28:12	0.95
  Optuna	40	0.97

  Optuna is taking more time and giving accuracy which is better than some models.