game3.py

import requests
import random
import time
import pandas as pd
import gradio as gr
import numpy as np
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import pipeline

def read3(num_selected_former):
    fname = 'data3_convai2_inferred.txt'
    with open(fname, encoding='utf-8') as f:
        content = f.readlines()
        index_selected = random.randint(0,len(content)/2-1)
        while index_selected == num_selected_former:
            index_selected = random.randint(0,len(content)/2-1)
        text = eval(content[index_selected*2])
        interpretation = eval(content[int(index_selected*2+1)])
        
        min_len = 5

        tokens = [i[0] for i in interpretation]
        tokens = tokens[1:-1]
        while len(tokens) <= min_len or '\\' in text['text'] or '//' in text['text']:
            index_selected = random.randint(0,len(content)/2-1)
            text = eval(content[int(index_selected*2)])
        res_tmp = [(i, 0) for i in text['text'].split(' ')]
        res = {"original": text['text'], "interpretation": res_tmp}    
    return res, index_selected
    
def func3(num_selected, human_predict, num1, num2, user_important):
    chatbot = []
    # num1: Human score; num2: AI score
    fname = 'data3_convai2_inferred.txt'

    with open(fname) as f:
        content = f.readlines()
        text = eval(content[int(num_selected*2)])
        interpretation = eval(content[int(num_selected*2+1)])
        
        if text['binary_label'] == 1:
            golden_label = int(50 * (1 - text['binary_score']))
        else:
            golden_label = int(50 * (1 + text['binary_score']))
            
    # (START) off-the-shelf version -- slow at the beginning
    # Load model directly
    # Use a pipeline as a high-level helper

    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
    classifier = pipeline("text-classification", model="padmajabfrl/Gender-Classification", device=device)
    output = classifier([text['text']])

    print(output)
    out = output[0]
    
    # (END) off-the-shelf version
    
    if out['label'] == 'Female':
        ai_predict = int(100 * out['score'])
    else:
        ai_predict = 100 - int(100 * out['score'])
    
    user_select = "You focused on "
    flag_select = False
    if user_important == "":
        user_select += "nothing. Interesting! "
    else:
        user_select += "'" + user_important + "'. "
    # for i in range(len(user_marks)):
    #     if user_marks[i][1] != None and h1[i][0] not in ["P", "N"]:
    #         flag_select = True
    #         user_select += "'" + h1[i][0] + "'"
    #         if i == len(h1) - 1:
    #             user_select += ". "
    #         else:
    #             user_select += ", "
    # if not flag_select:
    #     user_select += "nothing. Interesting! "
    user_select += "Wanna see how the AI made the guess? Click here. ⬅️"
    
    if golden_label > 60:
        gender = ' (female)'
    elif golden_label < 40:
        gender = ' (male)'    
    else:
        gender = ' (neutral)'
        
    if abs(golden_label - human_predict) <= 20  and abs(golden_label - ai_predict) <= 20:
        chatbot.append(("The correct answer is " + str(golden_label) + gender + ". Congratulations! 🎉 Both of you get the correct answer!", user_select))
        num1 += 1
        num2 += 1
    elif abs(golden_label - human_predict) > 20 and abs(golden_label - ai_predict) > 20:
        chatbot.append(("The correct answer is " + str(golden_label) + gender + ". Sorry.. No one gets the correct answer. But nice try! 😉", user_select))
    elif abs(golden_label - human_predict) <= 20 and abs(golden_label - ai_predict) > 20:
        chatbot.append(("The correct answer is " + str(golden_label) + gender + ". Great! 🎉 You are closer to the answer and better than AI!", user_select))
        num1 += 1
    else:
        chatbot.append(("The correct answer is " + str(golden_label) + gender + ". Sorry.. AI wins in this round.", user_select))
        num2 += 1
    
    tot_scores = ''' ### <p style="text-align: center;"> 🤖 Machine &ensp; ''' + str(int(num2)) + ''' &ensp; VS &ensp; ''' + str(int(num1)) + ''' &ensp; Human 👨👩 </p>'''

    
    num_tmp = max(num1, num2)
    y_lim_upper = (int((num_tmp + 3)/10)+1) * 10

    return ai_predict, chatbot, num1, num2, tot_scores

def interpre3(num_selected):
    fname = 'data3_convai2_inferred.txt'
    with open(fname) as f:
        content = f.readlines()
        text = eval(content[int(num_selected*2)])
        interpretation = eval(content[int(num_selected*2+1)])
    
    print(interpretation)

    res = {"original": text['text'], "interpretation": interpretation}
    # pos = []
    # neg = []
    # res = []
    # for i in interpretation:
    #     if i[1] > 0:
    #         pos.append(i[1])
    #     elif i[1] < 0:
    #         neg.append(i[1])
    #     else:
    #         continue
    # median_pos = np.median(pos)
    # median_neg = np.median(neg)


    # res.append(("P", "+"))
    # res.append(("/", None))
    # res.append(("N", "-"))
    # res.append(("Review:", None))
    # for i in interpretation:
    #     if i[1] > median_pos:
    #         res.append((i[0], "+"))
    #     elif i[1] < median_neg:
    #         res.append((i[0], "-"))
    #     else:
    #         res.append((i[0], None))
    return res

    
def func3_written(text_written, human_predict, lang_written):
    chatbot = []
    # num1: Human score; num2: AI score

    # (START) off-the-shelf version

    # tokenizer = AutoTokenizer.from_pretrained("nlptown/bert-base-multilingual-uncased-sentiment")
    # model = AutoModelForSequenceClassification.from_pretrained("nlptown/bert-base-multilingual-uncased-sentiment")

    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
    classifier = pipeline("text-classification", model="padmajabfrl/Gender-Classification", device=device)

    output = classifier([text_written])

    print(output)
    out = output[0]
    # (END) off-the-shelf version

    if out['label'] == 'Female':
        ai_predict = int(100 * out['score'])
    else:
        ai_predict = 100 - int(100 * out['score'])
    
    if abs(ai_predict - human_predict) <= 20:
        chatbot.append(("AI gives it a close score! 🎉", "⬅️ Feel free to try another one! ⬅️"))
    else:
        chatbot.append(("AI thinks in a different way from human. 😉", "⬅️ Feel free to try another one! ⬅️"))

    import shap

    gender_classifier = pipeline("text-classification", model="padmajabfrl/Gender-Classification", return_all_scores=True, device=device)

    explainer = shap.Explainer(gender_classifier)

    shap_values = explainer([text_written])
    interpretation = list(zip(shap_values.data[0], shap_values.values[0, :, 1]))
    
    res = {"original": text_written, "interpretation": interpretation}
    print(res)

    return res, ai_predict, chatbot