File size: 10,855 Bytes
48ae672
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a07131b
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
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
import gradio as gr
import torch
import numpy as np
import pandas as pd
import copy
import torch.nn.functional as F
from collections import defaultdict
from openprompt import PromptDataLoader, PromptForClassification
from openprompt.data_utils import InputExample
from openprompt.prompts import MixedTemplate, SoftVerbalizer
from transformers import AdamW, get_linear_schedule_with_warmup, XLMRobertaConfig, XLMRobertaTokenizer, XLMRobertaModel, XLMRobertaForMaskedLM, set_seed, AdapterConfig
from openprompt.plms.utils import TokenizerWrapper

import re

def check_only_numbers(string):
    return string.isdigit()

def remove_symbols_and_numbers(string):
    pattern = r"[-()\"#/@;:<>{}`+=~|_▁.!?,1234567890]"
    clean_string = re.sub(pattern, '', string)
    return clean_string

def is_sinhala(char):
  # https://unicode.org/charts/PDF/U0D80.pdf
  return ord(char) >= 0x0D80 and ord(char) <= 0x0DFF

def get_chars(word, without_si_modifiers = True):
  mods = [0x0DCA,0x0DCF,0x0DD0,0x0DD1,0x0DD2,0x0DD3,0x0DD4,0x0DD5,0x0DD6,0x0DD7,0x0DD8,0x0DD9,0x0DDA,0x0DDB,0x0DDC,0x0DDD,0x0DDE,0x0DDF,0x0DF2,0x0DF3]
  if without_si_modifiers:
    return [char for char in list(word) if ord(char) not in mods]
  else:
    return list(word)


def script_classify(text,en_thresh,si_thresh,without_si_mods):
    script = ""
    tokens = text.split()
    total_chars = 0
    latin_char_count = 0
    sin_char_count = 0
    for t_i,t in enumerate(tokens):    
        if check_only_numbers(t):
            continue
        token_list = get_chars(remove_symbols_and_numbers(t),without_si_modifiers = without_si_mods)
        token_len = len(token_list)
        total_chars += token_len
        for ch in token_list:
            if is_sinhala(ch):
                sin_char_count += 1
            else:
                latin_char_count += 1
    if total_chars == 0:
        script  = 'Symbol'
    else:
        en_percentage = latin_char_count/total_chars
        si_percentage = sin_char_count/total_chars
        if en_percentage >= en_thresh:
            script  = 'Latin'
        elif si_percentage >= si_thresh:
            script  = 'Sinhala'
        elif en_percentage < en_thresh and si_percentage < si_thresh:
            script = 'Mixed'
    return script

HUMOUR_MODEL_PATH = 'ad-houlsby-humour-seed-42.ckpt'
SENTIMENT_MODEL_PATH = 'ad-drop-houlsby-11-sentiment-seed-42.ckpt'
humour_mapping = {
    0: "Non-humourous",
    1:"Humourous"
}

sentiment_mapping = {
    0: "Negative",
    1:"Neutral",
    2:"Positive",
    3:"Conflict"
}

def load_plm(model_name, model_path):   
    model_config = XLMRobertaConfig.from_pretrained(model_path)
    model = XLMRobertaForMaskedLM.from_pretrained(model_path, config=model_config)
    tokenizer = XLMRobertaTokenizer.from_pretrained(model_path)
    wrapper = MLMTokenizerWrapper
    return model, tokenizer, wrapper

class MLMTokenizerWrapper(TokenizerWrapper):
    add_input_keys = ['input_ids', 'attention_mask', 'token_type_ids']

    @property
    def mask_token(self):
        return self.tokenizer.mask_token

    @property
    def mask_token_ids(self):
        return self.tokenizer.mask_token_id

    @property
    def num_special_tokens_to_add(self):
        if not hasattr(self, '_num_specials'):
            self._num_specials = self.tokenizer.num_special_tokens_to_add()
        return self._num_specials

    def tokenize_one_example(self, wrapped_example, teacher_forcing):
        wrapped_example, others = wrapped_example
        encoded_tgt_text = []
        if 'tgt_text' in others:
            tgt_text = others['tgt_text']
            if isinstance(tgt_text, str):
                tgt_text = [tgt_text]
            for t in tgt_text:
                encoded_tgt_text.append(self.tokenizer.encode(t, add_special_tokens=False))

        mask_id = 0 # the i-th the mask token in the template.

        encoder_inputs = defaultdict(list)
        for piece in wrapped_example:
            if piece['loss_ids']==1:
                if teacher_forcing: # fill the mask with the tgt task
                    raise RuntimeError("Masked Language Model can't perform teacher forcing training!")
                else:
                    encode_text = [self.mask_token_ids]
                mask_id += 1

            if piece['text'] in self.special_tokens_maps.keys():
                to_replace = self.special_tokens_maps[piece['text']]
                if to_replace is not None:
                    piece['text'] = to_replace
                else:
                    raise KeyError("This tokenizer doesn't specify {} token.".format(piece['text']))

            if 'soft_token_ids' in piece and piece['soft_token_ids']!=0:
                encode_text = [0] # can be replace by any token, since these token will use their own embeddings
            else:
                encode_text = self.tokenizer.encode(piece['text'], add_special_tokens=False)

            encoding_length = len(encode_text)
            encoder_inputs['input_ids'].append(encode_text)
            for key in piece:
                if key not in ['text']:
                    encoder_inputs[key].append([piece[key]]*encoding_length)

        encoder_inputs = self.truncate(encoder_inputs=encoder_inputs)
        # delete shortenable ids
        encoder_inputs.pop("shortenable_ids")
        encoder_inputs = self.concate_parts(input_dict=encoder_inputs)
        encoder_inputs = self.add_special_tokens(encoder_inputs=encoder_inputs)
        # create special input ids
        encoder_inputs['attention_mask'] = [1] *len(encoder_inputs['input_ids'])
        if self.create_token_type_ids:
            encoder_inputs['token_type_ids'] = [0] *len(encoder_inputs['input_ids'])
        # padding
        encoder_inputs = self.padding(input_dict=encoder_inputs, max_len=self.max_seq_length, pad_id_for_inputs=self.tokenizer.pad_token_id)

        if len(encoded_tgt_text) > 0:
            encoder_inputs = {**encoder_inputs, "encoded_tgt_text": encoded_tgt_text}# convert defaultdict to dict
        else:
            encoder_inputs = {**encoder_inputs}
        return encoder_inputs


plm, tokenizer, wrapper_class = load_plm("xlm", "xlm-roberta-base")
plm_copy = copy.deepcopy(plm)
tokenizer_copy = copy.deepcopy(tokenizer)
wrapper_class_copy = copy.deepcopy(wrapper_class)
sent_adapter_name = "Task_Sentiment"
sent_adapter_config = AdapterConfig.load("houlsby")
sent_adapter_config.leave_out.extend([11])
plm.add_adapter(sent_adapter_name, config=sent_adapter_config)
plm.set_active_adapters(sent_adapter_name)
plm.train_adapter(sent_adapter_name)
sent_template = '{"placeholder": "text_a"}. {"soft": "The"} {"soft": "sentiment"} {"soft": "or"} {"soft": "the"} {"soft": "feeling"} {"soft": "of"} {"soft": "the"} {"soft": "given"} {"soft": "sentence"} {"soft": "can"} {"soft": "be"} {"soft": "classified"} {"soft": "as"} {"soft": "positive"} {"soft": ","} {"soft": "negative"} {"soft": "or"} {"soft": "neutral"} {"soft": "."} {"soft": "The"} {"soft": "classified"} {"soft": "sentiment"} {"soft": "of"} {"soft": "the"} {"soft": "sentence"} {"soft": "is"} {"mask"}.'
sent_promptTemplate = MixedTemplate(model=plm, text = sent_template, tokenizer = tokenizer)
sent_promptVerbalizer = SoftVerbalizer(tokenizer, plm, num_classes=4)
sent_promptModel = PromptForClassification(template = sent_promptTemplate, plm = plm, verbalizer = sent_promptVerbalizer)
sent_promptModel.load_state_dict(torch.load(SENTIMENT_MODEL_PATH,map_location=torch.device('cpu')))
sent_promptModel.eval()

hum_adapter_name = "Ad_Humour"
hum_adapter_config = AdapterConfig.load("houlsby")
plm_copy.add_adapter(hum_adapter_name, config=hum_adapter_config)
plm_copy.set_active_adapters(hum_adapter_name)
plm_copy.train_adapter(hum_adapter_name)
hum_template = '{"placeholder": "text_a"}. {"soft": "Capture"} {"soft": "the"} {"soft": "comedic"} {"soft": "elements"} {"soft": "of"} {"soft": "the"} {"soft": "given"} {"soft": "sentence"} {"soft": "and"} {"soft": "classify"} {"soft": "as"} {"soft": "Humorous"} {"soft": ","} {"soft": "otherwise"} {"soft": "classify"} {"soft": "as"} {"soft": "Non-humorous"} {"soft": "."} {"soft": "The"} {"soft": "sentence"} {"soft": "is"} {"mask"}.'
hum_promptTemplate = MixedTemplate(model=plm_copy, text = hum_template, tokenizer = tokenizer_copy)
hum_promptVerbalizer = SoftVerbalizer(tokenizer_copy, plm_copy, num_classes=2)
hum_promptModel = PromptForClassification(template = hum_promptTemplate, plm = plm_copy, verbalizer = hum_promptVerbalizer)
hum_promptModel.load_state_dict(torch.load(HUMOUR_MODEL_PATH,map_location=torch.device('cpu')))
hum_promptModel.eval()

def sentiment(text):
    pred = None
    dataset = [
    InputExample(
        guid = 0,
        text_a = text,
    )
    ]
    data_loader = PromptDataLoader(
    dataset = dataset,
    tokenizer = tokenizer,
    template = sent_promptTemplate,
    tokenizer_wrapper_class=wrapper_class,
    )
    for step, inputs in enumerate(data_loader):
        logits = sent_promptModel(inputs)
        pred = sentiment_mapping[torch.argmax(logits, dim=-1).cpu().tolist()[0]]
    return pred    

def humour(text):
    pred = None
    dataset = [
    InputExample(
        guid = 0,
        text_a = text,
    )
    ]
    data_loader = PromptDataLoader(
    dataset = dataset,
    tokenizer = tokenizer_copy,
    template = hum_promptTemplate,
    tokenizer_wrapper_class=wrapper_class_copy,
    )
    for step, inputs in enumerate(data_loader):
        logits = hum_promptModel(inputs)
        pred = humour_mapping[torch.argmax(logits, dim=-1).cpu().tolist()[0]]
    return pred        


def classifier(text, task):
    one_script = script_classify(text,1.0,1.0,True)
    pointnine_script = script_classify(text,0.9,0.9,True)
    if task == "Sentiment Classification":
        return sentiment(text),one_script, pointnine_script
    elif task == "Humour Detection":
        return humour(text),one_script, pointnine_script


demo = gr.Interface(
    title="Use of Prompt-Based Learning For Code-Mixed Text Classification",
    fn=classifier, 
    inputs=[
        gr.Textbox(placeholder="Enter an input sentence...",label="Input Sentence"),
        gr.Radio(["Sentiment Classification", "Humour Detection"], label="Task")
    ], 
    outputs=[
        gr.Label(label="Label"),
        gr.Textbox(label="Script Threshold 100%"),
        gr.Textbox(label="Script Threshold 90%")
    ],
    allow_flagging = "never",
    examples=[
        ["Mama kamathi cricket matches balanna", "Sentiment Classification"],
        ["මම sweet food වලට කැමති නෑ", "Sentiment Classification"],
        ["The weather outside is neither too hot nor too cold", "Sentiment Classification"],
        ["ඉබ්බයි හාවයි හොඳ යාලුවොලු", "Humour Detection"],
        ["Kandy ගොඩක් lassanai", "Humour Detection"]
    ])

demo.launch()