allpreds.py

from gcode import predict_text_classification_single_label_sample
import pandas as pd
from dotenv import load_dotenv
import os
from quickchart import QuickChart

load_dotenv()

# this dictionary contains the kinds of bias and the Vertex endpoint IDs 
list_of_biases_and_endpts = {"Gender Bias": {"bias_type": "gender_bias", "endpoint_id": os.environ.get("GENDER_ENDPOINT_ID")},
                             "Racial Bias": {"bias_type": "racial_bias", "endpoint_id": os.environ.get("RACIAL_ENDPOINT_ID")},
                             "Political Bias": {"bias_type": "political_bias", "endpoint_id": os.environ.get("POLITICAL_ENDPOINT_ID")},
                             "Hate Speech": {"bias_type": "hate_speech", "endpoint_id": os.environ.get("HATE_ENDPOINT_ID")}}

# this dictionary keeps track of the order of biased confidence score
# (if order = 1, it means that at index 1 the value is bias confidence, if order =0 it means that at index 0 the value is bias confidence)
order_in_confidence = {"gender_bias": 0, "racial_bias": 1, "political_bias": 0, "hate_speech": 0}



# make_preds is like the boss of all functions and it uses predict function and generateChart function to do smaller tasks
def make_preds(content, bias_type):
    pos_tokens=[] # this stores the list of tuples which are passed to the highlightText component
    bias_scores={} # this dictionary is for the bar chart for the Bias %
    bias = list_of_biases_and_endpts[bias_type]
    df, bias_percentage = predict(content, bias["bias_type"], bias["endpoint_id"])
    bias_scores[bias_type] = bias_percentage
    # these lines of code are responsible for generating the data for highlightText component
    for ind in df.index:
        pos_tokens.extend([(df['content'][ind], bias_type if df['predictions'][ind]>0.5 else None), (" ", None)])
    
    # The next line makes use of a tool called quickcharts to create a radial guage chart
    html = generateChart(bias_percentage)

    # the three values returned here are used by the UI output components
    return pos_tokens, bias_scores, html



# the predict function acutally generates the predictions for the user content and it returns a dataframe containing
# the 'content' and 'prediction' columns. The prediction column contains the bias confidence score. 
# predict function also returns the bias percentage
def predict(content, bias_type, endpoint_id):
    # split the article into sentences
    chunks = split_into_sentences(content)
    possibly_biased = []
    # define the dataframe with two columns - 'content' and 'predictions'
    df = pd.DataFrame(columns=['content', 'predictions'])
    # for each chunk in the content, create a prediction and add a row to the dataframe
    for content in chunks:
        predictions = predict_text_classification_single_label_sample(
            project=os.environ.get("PROJECT_ID") ,
            endpoint_id=endpoint_id,
            location="us-central1",
            content=content
        )
        for prediction in predictions:
            res=float(dict(prediction)['confidences'][order_in_confidence[bias_type]])
            new_row  = {'content': content, 'predictions': res}
            df.loc[len(df)] =  new_row
            if dict(prediction)['confidences'][order_in_confidence[bias_type]] > 0.5:
                possibly_biased.append(content)
        
    # save the dataframe as a csv file
    df.to_csv(f'preds_{bias_type}.csv')
    bias_percentage = round(len(possibly_biased)/len(chunks), 2)
    return df, bias_percentage

# this function splits the content into 20 word chunks
def split_into_20_word_chunks(long_string):
    words = long_string.split()
    chunks = []
    chunk_size = 20

    for i in range(0, len(words), chunk_size):
        chunk = ' '.join(words[i:i + chunk_size])
        chunks.append(chunk)

    return chunks

# this function splits the content into sentences
def split_into_sentences(long_string):
    sentences = []
    current_sentence = ""
    punctuation_marks = {'.', '?', '!'}

    for char in long_string:
        current_sentence += char
        if char in punctuation_marks:
            sentences.append(current_sentence.strip())
            current_sentence = ""

    if current_sentence:
        sentences.append(current_sentence.strip())

    return sentences


# generatesChart function creates the circular bias Percentage chart
# it uses the quickChart library which is used to plot charts and graphs
def generateChart(bias_percentage):
    qc = QuickChart()
    qc.width = 500
    qc.height = 300
    qc.version = '2'

    # Config can be set as a string or as a nested dict
    qc.config = """{
    type: 'radialGauge',
    data: {
        datasets: [{
        data: ["""+str(round(bias_percentage*100, 0))+"""],
        backgroundColor: getGradientFillHelper('horizontal', ['red', 'blue']),
        }]
    },
    options: {
        // See https://github.com/pandameister/chartjs-chart-radial-gauge#options
        domain: [0, 100],
        trackColor: '#f0f8ff', 
        centerPercentage: 90,
        centerArea: {
        text: (val) => val + '%',
        },
    }
    }"""
        
    url=qc.get_url()
    html=f"""<img src="{url}"/>"""
    html = (
    "<div style='max-width:100%; max-height:360px; overflow:auto'>"
    + html
    + "</div>")
    return html