Simple CLI for analyzing Google's Bert predicted data.

bert-text is designed to allow simple statistics measurements of binary classifier's output preditions. By deafult Bert created the predicted classes columns by their order of appearence in the train.tsv file, bert-text takes care of remembering, how the values were presented and uses this information to evaluate the predicted data.

bert-analysis uses npyscreen to create command line interface

pip install npyscreen

1. Traning file(train.tsv) This is the file used while traning Bert, the file is needed to analyze and remember the classes order of appearence.

Example

2. Testing file This is the file where you keep the classes for data used for testing bert.This file is needed because bert doesn't take the actual class for it's traning data, so the file is there for comparison.

Example

2. Predicted file(test_results.tsv) This is the file returned by bert after prediction. This file is needed to test the accuracy of bert output.

Example

• True Positives
• False Positives
• False Negatives
• False Positives
• Accuracu
• Specificity
• Sensitivity
• Precision
• F-score
• Matthews correlation coefficient
• Youden Index

bert-analysis is currently limited to analyzing binary classifiers, this might change in future.

• Confussion martix as image
• ROC curve
• AUC
• Multinomial classifiers
• Returning hard sentences