Define target (test data + metric)

[arjunjauhari]

1) Split data (train/dev/test) 2) Define a metric (code it)

[naik-amey]

Metric:

1. The accuracy of prediction. Prediction being if the pair is duplicate. I think this is the only metric.

[naik-amey]

Summary: Total number of question pairs for training: 404290 Duplicate pairs: 36.92% Total number of questions in the training data: 537933 Number of questions that appear multiple times: 111780 Total number of question pairs for testing: 2345796

[naik-amey]

Metrics to evaluate:

1. accuracy
2. confusion matrix
3. precision
4. recall
5. sensitivity
6. specificity
7. F1 score
8. ROC
9. AUC

[naik-amey]

Step 1: Extract a feature Step 2: Apply logistic regression.