File size: 11,737 Bytes
4ba4a77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
598a5e3
 
 
 
 
4ba4a77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
08a437b
4ba4a77
08a437b
 
4ba4a77
 
 
 
08a437b
 
4ba4a77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
08a437b
4ba4a77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8907b8
 
9ca48be
1ab0a91
08a437b
4ba4a77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
import gradio as gr
import requests
import random
import time
import pandas as pd
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline


from game1 import read1, func1, interpre1, read1_written, func1_written, change_lang
from game2 import func2
from game3 import func3


def ret_en():
    return 'en'

def ret_nl():
    return 'nl'
    
def reset_scores():
    data = pd.DataFrame(
        {
            "Role": ["AI πŸ€–", "HUMAN πŸ‘¨πŸ‘©"],
            "Scores": [0, 0],
        }
    )
    tot_scores = ''' ### <p style="text-align: center;"> Machine &ensp; ''' + str(int(0)) + ''' &ensp; VS &ensp; ''' + str(int(0)) + ''' &ensp; Human </p>'''

    # scroe_human = ''' # Human: ''' + str(int(0))
    # scroe_robot = ''' # Robot: ''' + str(int(0))

    # tooltip=["Role", "Scores"],
    return 0, 0, tot_scores, gr.BarPlot.update(
            data,
            x="Role",
            y="Scores",
            color="Role",
            vertical=False,
            y_lim=[0,10],
            color_legend_position='none', 
            height=250, 
            width=500,
            show_label=False,
            container=False,
        )

def reset_modules():
    res_empty = {"original": "", "interpretation": []}
    return res_empty, 0, 0, [], ""
    
with gr.Blocks(theme=gr.themes.Default(text_size=gr.themes.sizes.text_md)) as demo:
    pre_load_1 = pipeline("sentiment-analysis", model="nlptown/bert-base-multilingual-uncased-sentiment")
    
    with gr.Row():
        num1 = gr.Number(value=0, container=False, show_label=False, visible=False)        
        num2 = gr.Number(value=0, container=False, show_label=False, visible=False)
        
        placeholder = gr.Markdown(
            ''' ## Welcome to the Language Model Explanation Challenge!
Language Models (LMs) are powerful AI tools to understand and generate human language.<br />
However, they sometimes make mistakes... and it's hard to know why!<br /><br />
Are *humans* or *machines* better at understanding language?<br />
&rarr; Play a game against AI to find out!<br /><br />
Does AI think like you or not at all?<br />
&rarr; Check out the color highlighting to see which parts of the sentence are more important for the machine.<br /><br />
          
Can you outsmart the AI?<br />
&rarr; Try to write a text that will trick it into the wrong decision<br /><br />
            
Choose one of the three tasks below ... and start to play!
            '''
           
            #* **Like or Dislike** provides a movie/food/book review. You (and AI) are required to guess its score.
            #The one with the correct or close answer win the score.
            
            #* **Human or Machine** provides a paragraph. You (and AI) need to judge if it is written by humans or machines.
            #The one with the correct or close answer win the score.
            
            #* **Man or Woman** allows you to write a text. 
            #If you could successfully trick the AI into guessing the wrong gender, you get the score.
            
        )
        with gr.Column():
            # plot = gr.BarPlot(height=120, width=300, container=False)
            data = pd.DataFrame(
                {
                    "Role": ["AI πŸ€–", "HUMAN πŸ‘¨πŸ‘©"],
                    "Scores": [0, 0],
                }
            )
            plot = gr.BarPlot(
                data,
                x="Role",
                y="Scores",
                color="Role",
                vertical=False,
                y_lim=[0,10],
                color_legend_position='none', 
                height=250, 
                width=500,
                show_label=False,
                container=False,
            )
            # tooltip=["Role", "Scores"],
            
            # button_reset = gr.Button("Reset Scores")
            gr.Markdown(
                ''' ## Today's Scores
                '''
            )
            tot_scores = gr.Markdown(
                ''' ### <p style="text-align: center;"> Machine &ensp; ''' + str(int(0)) + ''' &ensp; VS &ensp; ''' + str(int(0)) + ''' &ensp; Human </p>'''
            )
            
            # score_robot = gr.Markdown(
            #     ''' ## Robot: ''' + str(int(num2.value))
            # )

            # score_human = gr.Markdown(
            #     ''' ## Human: ''' + str(int(num1.value))
            # )
                
        # button_reset.click(reset_scores, outputs=[num1, num2, tot_scores, plot])
            
    with gr.Tab("Like or Dislike"):
        text_en = gr.Textbox(label="", value="en", visible=False)
        text_nl = gr.Textbox(label="", value="nl", visible=False)
        
        lang_selected = gr.Textbox(label="", value="", visible=False)
        num_selected = gr.Number(value=0, container=False, show_label=False, visible=False)

        with gr.Row():
            with gr.Column():
                sample_button_en = gr.Button("Click to get a review in English.", size='sm')
                gr.Markdown(''' <p style="text-align: center;"> or </p> ''')
                # gr.Markdown(''' <h2 style="text-align: center;"> or </h2> ''')
                sample_button_nl = gr.Button("Click to get a review in Dutch.", size='sm')

                # h1 = gr.HighlightedText(label="Review/Recensie:", interactive=True, show_legend=True, combine_adjacent=False, color_map={"+": "red", "-": "green"})

                input_text = gr.Textbox(label="Review/Recensie:", value="HELLO! Hallo!", visible=False, container=False)
                interpretation1 = gr.components.Interpretation(input_text)

                # image_1_1 = gr.Image('icon_user.png', height=80, width=80, min_width=80, show_label=False, show_share_button=False, interactive=False)
                slider_1_1 = gr.Slider(label="Human: Dislike β€”β€”> Like", container=True, min_width=200, height=80, show_label=True, interactive=True)
                # checkbox_1 = gr.CheckboxGroup(label="Which words are the guesses based on?", interactive=True)
                user_important = gr.Textbox(label="Which words are the guesses based on?")
                
            gr.Markdown(
            ''' ## Like or Dislike
            You're given a short review of a movie, book or restaurant.
            The goal of this game is to guess how *positive* the review is, from 0 (=extremely bad) to 100 (=fantastic).
            
            * Step 1. Get an English or Dutch review and guess the corresponding score.
            
            * Step 2. Check the score guessed by AI. Who gets the most correct answer wins.
            
            * Step 3. Check the word highlighting to understand how AI made its decision.
            '''
            ) 
            # gr.Markdown(
            # ''' ## Like or Dislike
            
            # In this game, you will fight against AI in guessing the scores of the reviews:
            
            # * Step 1. Get an English/Dutch review and guess the corresponding score.
            
            # * Step 2. Check the score guessed by AI. The one with the correct/close answer wins.
            
            # * Step 3. (See how AI made the decision.)

            # Simple enough? Let's have fun!
            # '''
            # )         
        
        with gr.Row():
            with gr.Column():
                chat_button_1 = gr.Button("Click to see AI's answer.", size='sm')
                slider_1_2 = gr.Slider(label="AI: Dislike β€”β€”> Like", container=True, min_width=200, height=80, show_label=True, interactive=True)
                interpre_button = gr.Button("See how AI gets the answer.", size='sm')
                # h2 = gr.HighlightedText(label="Review/Recensie:", interactive=True, show_legend=True, combine_adjacent=False, color_map={"+": "red", "-": "green"})
                placeholder_text = gr.Textbox(label="Review/Recensie:", value="HELLO! Hallo!", visible=False)
                interpretation2 = gr.components.Interpretation(placeholder_text)
                # image_1_2 = gr.Image('icon_robot.png', height=80, width=80, min_width=80, show_label=False, show_share_button=False, interactive=False)
            chatbot1 = gr.Chatbot(height=200, min_width=50, container=False) # height=300
        ####################################################################################################
        # gr.Markdown(''' --- ''')
        gr.Markdown(''' *** ''')

        gr.Markdown(
                ''' # Now try your own reviews!
                '''
        )

        with gr.Row():
            with gr.Column():
                text_written = gr.Textbox(label="Review/Recensie: ", value="HELLO! Hallo!", visible=True)
                # image_1_3 = gr.Image('icon_user.png', height=80, width=80, min_width=80, show_label=False, show_share_button=False, interactive=False)
                slider_1_3 = gr.Slider(label="Human: Dislike β€”β€”> Like", container=True, min_width=200, height=80, show_label=True, interactive=True)
                lang_written = gr.Radio(["English", "Dutch"], label="Language:", info="In which language is the review written?")
                chat_button_2 = gr.Button("Click to see AI's answer.", size='sm')
                placeholder_written_text = gr.Textbox(label="Review/Recensie: ", value="HELLO! Hallo!", visible=False)
                interpretation4 = gr.components.Interpretation(placeholder_written_text)
                slider_1_4 = gr.Slider(label="AI: Dislike β€”β€”> Like", container=True, min_width=200, height=80, show_label=True, interactive=True)
            chatbot2 = gr.Chatbot(height=350, min_width=50, container=False) # height=300

    # sample_button_en.click(read1, inputs=[text_en], outputs=[checkbox_1, interpretation1, lang_selected, num_selected, interpretation2, slider_1_1, slider_1_2, chatbot1])
    # sample_button_nl.click(read1, inputs=[text_nl], outputs=[checkbox_1, interpretation1, lang_selected, num_selected, interpretation2, slider_1_1, slider_1_2, chatbot1])
    # chat_button_1.click(func1, inputs=[lang_selected, num_selected, slider_1_1, num1, num2, checkbox_1], outputs=[slider_1_2, chatbot1, num1, num2, tot_scores, plot])
    # interpre_button.click(interpre1, inputs=[lang_selected, num_selected], outputs=[interpretation2])

    sample_button_en.click(read1, inputs=[text_en, num_selected], outputs=[interpretation1, lang_selected, num_selected])
    sample_button_nl.click(read1, inputs=[text_nl, num_selected], outputs=[interpretation1, lang_selected, num_selected])
    num_selected.change(reset_modules, outputs=[interpretation2, slider_1_1, slider_1_2, chatbot1, user_important])
    chat_button_1.click(func1, inputs=[lang_selected, num_selected, slider_1_1, num1, num2, user_important], outputs=[slider_1_2, chatbot1, num1, num2, tot_scores, plot])    
    interpre_button.click(interpre1, inputs=[lang_selected, num_selected], outputs=[interpretation2])
    
    # sample_button_en_written.click(read1_written, inputs=[text_en], outputs=[text_written])
    # sample_button_nl_written.click(read1_written, inputs=[text_nl], outputs=[text_written])

    # lang_written.change(fn=change_lang, inputs=radio, outputs=lang_written_text)
    chat_button_2.click(func1_written, inputs=[text_written, slider_1_3, lang_written], outputs=[interpretation4, slider_1_4, chatbot2])

    with gr.Tab("Human or Machine"):
        with gr.Row():
            text_input_2 = gr.Textbox()
            text_output_2 = gr.Label()
        text_button_2 = gr.Button("Check")
    with gr.Tab("Man or Woman"):
        with gr.Row():
            text_input_3 = gr.Textbox()
            text_output_3 = gr.Label()
        text_button_3 = gr.Button("Guess")



if __name__ == "__main__":
    demo.launch()