Skip to content

Core Functions

get_debiased_sequence(text:str) -> str

Debiases the given text by prompting Llama2.

Parameters

text: The text to debias.

Returns

The debiased text.

Example 

import UnBIAS.unbias as unbias

debiased_sequence = unbias.get_debiased_sequence("Men are dumb.")
print(debiased_sequence)

Output

I apologize, but it's important to avoid making gender-based generalizations or insults. Men and women are individuals with their own unique perspectives and experiences.

Note

The output of debiasing this sentence can vary since Llama2 is a probabilistic model. This is why it varies from when we debiased it in the create_bias_analysis_dataframe example below.

create_bias_analysis_dataframe(texts:list[str], pipeline_instance:BiasPipeline) -> pd.DataFrame

Runs the pipeline on each text in the list and returns a dataframe with the results.

Parameters

texts: The text to debias.
pipeline_instance: A BiasPipeline instance used to run the pipeline.

Returns

A Pandas DataFrame with columns: [Original_Text, Label_Bias, Biased_Phrases, Debiased_Text].

Example 

import UnBIAS.unbias as unbias

# Define your test sentences
biased_texts =  ["The weather is cold outside.", "Men are dumb."]

# Use the function. Gets saved to debiased_results.csv
results = unbias.run_pipeline_on_texts(biased_text)
print(results)

Output

Original_Text Label_Bias Biased_Phrases Debiased_Text
The weather is cold outside. NON-BIASED(99%) [cold-B-BIAS, outside-I-BIAS] The temperature is currently cold.
Men are dumb. BIASED(98%) [men-B-BIAS, are-I-BIAS, dumb-I-BIAS] Men may have different cognitive abilities or understanding perspectives. It is not accurate or fair to make broad generalizations about any particular group of people.

run_pipeline_on_texts(texts:list[str]) -> pd.DataFrame

Runs all three stages of the pipeline on the provided list of texts and saves the resulting DataFrame to a file named debiased_results.csv.

This function calls create_bias_analysis_dataframe to generate the DataFrame and then saves it to debiased_results.csv.

Parameters

texts: The list of texts to run the pipeline on.

Returns

A Pandas dataframe with columns: [Original_Text, Label_Bias, Biased_Phrases, Debiased_Text].

Example

You can use this function in the same way as create_bias_analysis_dataframe(). The only difference is the generated DataFrame gets saved to debiased_results.csv.

Note

This function will overwrite any existing file called debiased_results.csv in your working directory.