Upload 3 files
Browse files- app.py +132 -0
- infer.ipynb +171 -0
- requirements.txt +3 -0
app.py
ADDED
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import torch
|
3 |
+
from tqdm import tqdm
|
4 |
+
from torch.utils.data import Dataset, DataLoader
|
5 |
+
from transformers import DistilBertTokenizer, DistilBertModel
|
6 |
+
import os
|
7 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
|
8 |
+
|
9 |
+
|
10 |
+
|
11 |
+
MAX_LEN = 512
|
12 |
+
TRAIN_BATCH_SIZE = 16
|
13 |
+
VALID_BATCH_SIZE = 16
|
14 |
+
EPOCHS = 3
|
15 |
+
LEARNING_RATE = 1e-05
|
16 |
+
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
17 |
+
|
18 |
+
|
19 |
+
tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased', truncation=True, do_lower_case=True)
|
20 |
+
|
21 |
+
class MultiLabelDataset(Dataset):
|
22 |
+
|
23 |
+
def __init__(self, dataframe, tokenizer, max_len, new_data=False):
|
24 |
+
self.tokenizer = tokenizer
|
25 |
+
self.data = dataframe
|
26 |
+
self.text = dataframe.comment_text
|
27 |
+
self.new_data = new_data
|
28 |
+
|
29 |
+
if not new_data:
|
30 |
+
self.targets = self.data.labels
|
31 |
+
self.max_len = max_len
|
32 |
+
|
33 |
+
def __len__(self):
|
34 |
+
return len(self.text)
|
35 |
+
|
36 |
+
def __getitem__(self, index):
|
37 |
+
text = str(self.text[index])
|
38 |
+
text = " ".join(text.split())
|
39 |
+
|
40 |
+
inputs = self.tokenizer.encode_plus(
|
41 |
+
text,
|
42 |
+
None,
|
43 |
+
add_special_tokens=True,
|
44 |
+
max_length=self.max_len,
|
45 |
+
pad_to_max_length=True,
|
46 |
+
return_token_type_ids=True
|
47 |
+
)
|
48 |
+
ids = inputs['input_ids']
|
49 |
+
mask = inputs['attention_mask']
|
50 |
+
token_type_ids = inputs["token_type_ids"]
|
51 |
+
|
52 |
+
out = {
|
53 |
+
'ids': torch.tensor(ids, dtype=torch.long),
|
54 |
+
'mask': torch.tensor(mask, dtype=torch.long),
|
55 |
+
'token_type_ids': torch.tensor(token_type_ids, dtype=torch.long),
|
56 |
+
}
|
57 |
+
|
58 |
+
if not self.new_data:
|
59 |
+
out['targets'] = torch.tensor(self.targets[index], dtype=torch.float)
|
60 |
+
|
61 |
+
return out
|
62 |
+
|
63 |
+
class DistilBERTClass(torch.nn.Module):
|
64 |
+
def __init__(self):
|
65 |
+
super(DistilBERTClass, self).__init__()
|
66 |
+
|
67 |
+
self.bert = DistilBertModel.from_pretrained("distilbert-base-uncased")
|
68 |
+
self.classifier = torch.nn.Sequential(
|
69 |
+
torch.nn.Linear(768, 768),
|
70 |
+
torch.nn.ReLU(),
|
71 |
+
torch.nn.Dropout(0.1),
|
72 |
+
torch.nn.Linear(768, 6)
|
73 |
+
)
|
74 |
+
|
75 |
+
def forward(self, input_ids, attention_mask, token_type_ids):
|
76 |
+
output_1 = self.bert(input_ids=input_ids, attention_mask=attention_mask)
|
77 |
+
hidden_state = output_1[0]
|
78 |
+
out = hidden_state[:, 0]
|
79 |
+
out = self.classifier(out)
|
80 |
+
return out
|
81 |
+
|
82 |
+
model = DistilBERTClass()
|
83 |
+
model.to(DEVICE);
|
84 |
+
|
85 |
+
model_loaded = torch.load('model/inference_models_output_4fold_distilbert_fold_best_model.pth')
|
86 |
+
|
87 |
+
model.load_state_dict(model_loaded['model'])
|
88 |
+
|
89 |
+
|
90 |
+
val_params = {'batch_size': VALID_BATCH_SIZE,
|
91 |
+
'shuffle': False,
|
92 |
+
|
93 |
+
}
|
94 |
+
def give_toxic(text):
|
95 |
+
# text = "You fucker "
|
96 |
+
test_data = pd.DataFrame([text],columns=['comment_text'])
|
97 |
+
test_set = MultiLabelDataset(test_data, tokenizer, MAX_LEN, new_data=True)
|
98 |
+
test_loader = DataLoader(test_set, **val_params)
|
99 |
+
|
100 |
+
all_test_pred = []
|
101 |
+
|
102 |
+
def test(epoch):
|
103 |
+
model.eval()
|
104 |
+
|
105 |
+
with torch.inference_mode():
|
106 |
+
|
107 |
+
for _, data in tqdm(enumerate(test_loader, 0)):
|
108 |
+
|
109 |
+
|
110 |
+
ids = data['ids'].to(DEVICE, dtype=torch.long)
|
111 |
+
mask = data['mask'].to(DEVICE, dtype=torch.long)
|
112 |
+
token_type_ids = data['token_type_ids'].to(DEVICE, dtype=torch.long)
|
113 |
+
outputs = model(ids, mask, token_type_ids)
|
114 |
+
probas = torch.sigmoid(outputs)
|
115 |
+
|
116 |
+
all_test_pred.append(probas)
|
117 |
+
|
118 |
+
probas = test(model)
|
119 |
+
|
120 |
+
all_test_pred = torch.cat(all_test_pred)
|
121 |
+
|
122 |
+
label_columns = ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"]
|
123 |
+
|
124 |
+
preds = all_test_pred.detach().cpu().numpy()[0]
|
125 |
+
|
126 |
+
final_dict = dict(zip(label_columns , preds))
|
127 |
+
return final_dict
|
128 |
+
|
129 |
+
def device():
|
130 |
+
return DEVICE
|
131 |
+
|
132 |
+
print(give_toxic("fuck"))
|
infer.ipynb
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cells": [
|
3 |
+
{
|
4 |
+
"cell_type": "code",
|
5 |
+
"execution_count": 20,
|
6 |
+
"id": "d136f503-bb1b-404e-8657-ce3168eae54b",
|
7 |
+
"metadata": {},
|
8 |
+
"outputs": [],
|
9 |
+
"source": [
|
10 |
+
"import pandas as pd\n",
|
11 |
+
"import torch\n",
|
12 |
+
"from tqdm import tqdm\n",
|
13 |
+
"from torch.utils.data import Dataset, DataLoader, RandomSampler, SequentialSampler\n",
|
14 |
+
"from transformers import DistilBertTokenizer, DistilBertModel\n",
|
15 |
+
"import streamlit as st\n",
|
16 |
+
"\n",
|
17 |
+
"\n",
|
18 |
+
"\n",
|
19 |
+
"\n",
|
20 |
+
"MAX_LEN = 512\n",
|
21 |
+
"TRAIN_BATCH_SIZE = 16\n",
|
22 |
+
"VALID_BATCH_SIZE = 16\n",
|
23 |
+
"EPOCHS = 3\n",
|
24 |
+
"LEARNING_RATE = 1e-05\n",
|
25 |
+
"DEVICE = 'cuda:0' if torch.cuda.is_available() else 'cpu'\n",
|
26 |
+
"print(DEVICE)\n",
|
27 |
+
"\n",
|
28 |
+
"tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased', truncation=True, do_lower_case=True)\n",
|
29 |
+
"\n",
|
30 |
+
"class MultiLabelDataset(Dataset):\n",
|
31 |
+
"\n",
|
32 |
+
" def __init__(self, dataframe, tokenizer, max_len, new_data=False):\n",
|
33 |
+
" self.tokenizer = tokenizer\n",
|
34 |
+
" self.data = dataframe\n",
|
35 |
+
" self.text = dataframe.comment_text\n",
|
36 |
+
" self.new_data = new_data\n",
|
37 |
+
" \n",
|
38 |
+
" if not new_data:\n",
|
39 |
+
" self.targets = self.data.labels\n",
|
40 |
+
" self.max_len = max_len\n",
|
41 |
+
"\n",
|
42 |
+
" def __len__(self):\n",
|
43 |
+
" return len(self.text)\n",
|
44 |
+
"\n",
|
45 |
+
" def __getitem__(self, index):\n",
|
46 |
+
" text = str(self.text[index])\n",
|
47 |
+
" text = \" \".join(text.split())\n",
|
48 |
+
"\n",
|
49 |
+
" inputs = self.tokenizer.encode_plus(\n",
|
50 |
+
" text,\n",
|
51 |
+
" None,\n",
|
52 |
+
" add_special_tokens=True,\n",
|
53 |
+
" max_length=self.max_len,\n",
|
54 |
+
" pad_to_max_length=True,\n",
|
55 |
+
" return_token_type_ids=True\n",
|
56 |
+
" )\n",
|
57 |
+
" ids = inputs['input_ids']\n",
|
58 |
+
" mask = inputs['attention_mask']\n",
|
59 |
+
" token_type_ids = inputs[\"token_type_ids\"]\n",
|
60 |
+
"\n",
|
61 |
+
" out = {\n",
|
62 |
+
" 'ids': torch.tensor(ids, dtype=torch.long),\n",
|
63 |
+
" 'mask': torch.tensor(mask, dtype=torch.long),\n",
|
64 |
+
" 'token_type_ids': torch.tensor(token_type_ids, dtype=torch.long),\n",
|
65 |
+
" }\n",
|
66 |
+
" \n",
|
67 |
+
" if not self.new_data:\n",
|
68 |
+
" out['targets'] = torch.tensor(self.targets[index], dtype=torch.float)\n",
|
69 |
+
"\n",
|
70 |
+
" return out\n",
|
71 |
+
"\n",
|
72 |
+
"class DistilBERTClass(torch.nn.Module):\n",
|
73 |
+
" def __init__(self):\n",
|
74 |
+
" super(DistilBERTClass, self).__init__()\n",
|
75 |
+
" \n",
|
76 |
+
" self.bert = DistilBertModel.from_pretrained(\"distilbert-base-uncased\")\n",
|
77 |
+
" self.classifier = torch.nn.Sequential(\n",
|
78 |
+
" torch.nn.Linear(768, 768),\n",
|
79 |
+
" torch.nn.ReLU(),\n",
|
80 |
+
" torch.nn.Dropout(0.1),\n",
|
81 |
+
" torch.nn.Linear(768, 6)\n",
|
82 |
+
" )\n",
|
83 |
+
"\n",
|
84 |
+
" def forward(self, input_ids, attention_mask, token_type_ids):\n",
|
85 |
+
" output_1 = self.bert(input_ids=input_ids, attention_mask=attention_mask)\n",
|
86 |
+
" hidden_state = output_1[0]\n",
|
87 |
+
" out = hidden_state[:, 0]\n",
|
88 |
+
" out = self.classifier(out)\n",
|
89 |
+
" return out\n",
|
90 |
+
"\n",
|
91 |
+
"model = DistilBERTClass()\n",
|
92 |
+
"model.to(DEVICE);\n",
|
93 |
+
"\n",
|
94 |
+
"model_loaded = torch.load('model/inference_models_output_4fold_distilbert_fold_best_model.pth',map_location=torch.device('cpu'))\n",
|
95 |
+
"\n",
|
96 |
+
"model.load_state_dict(model_loadede['model'])\n",
|
97 |
+
"\n",
|
98 |
+
"\n",
|
99 |
+
"val_params = {'batch_size': VALID_BATCH_SIZE,\n",
|
100 |
+
" 'shuffle': False,\n",
|
101 |
+
" 'num_workers': 8\n",
|
102 |
+
" }\n",
|
103 |
+
"def give_toxic(text):\n",
|
104 |
+
" text = \"You fucker \"\n",
|
105 |
+
" test_data = pd.DataFrame([text],columns=['comment_text'])\n",
|
106 |
+
" test_set = MultiLabelDataset(test_data, tokenizer, MAX_LEN, new_data=True)\n",
|
107 |
+
" test_loader = DataLoader(test_set, **val_params)\n",
|
108 |
+
"\n",
|
109 |
+
" all_test_pred = []\n",
|
110 |
+
"\n",
|
111 |
+
" def test(epoch):\n",
|
112 |
+
" model.eval()\n",
|
113 |
+
"\n",
|
114 |
+
" with torch.inference_mode():\n",
|
115 |
+
"\n",
|
116 |
+
" for _, data in tqdm(enumerate(test_loader, 0)):\n",
|
117 |
+
"\n",
|
118 |
+
"\n",
|
119 |
+
" ids = data['ids'].to(DEVICE, dtype=torch.long)\n",
|
120 |
+
" mask = data['mask'].to(DEVICE, dtype=torch.long)\n",
|
121 |
+
" token_type_ids = data['token_type_ids'].to(DEVICE, dtype=torch.long)\n",
|
122 |
+
" outputs = model(ids, mask, token_type_ids)\n",
|
123 |
+
" probas = torch.sigmoid(outputs)\n",
|
124 |
+
"\n",
|
125 |
+
" all_test_pred.append(probas)\n",
|
126 |
+
"\n",
|
127 |
+
"\n",
|
128 |
+
" probas = test(model)\n",
|
129 |
+
"\n",
|
130 |
+
" all_test_pred = torch.cat(all_test_pred)\n",
|
131 |
+
"\n",
|
132 |
+
" label_columns = [\"toxic\", \"severe_toxic\", \"obscene\", \"threat\", \"insult\", \"identity_hate\"]\n",
|
133 |
+
"\n",
|
134 |
+
" preds = all_test_pred.detach().cpu().numpy()[0]\n",
|
135 |
+
"\n",
|
136 |
+
" final_dict = dict(zip(label_columns , preds))\n",
|
137 |
+
" return final_dict\n",
|
138 |
+
"\n"
|
139 |
+
]
|
140 |
+
},
|
141 |
+
{
|
142 |
+
"cell_type": "code",
|
143 |
+
"execution_count": null,
|
144 |
+
"id": "db651873-60cd-4cd7-8ba0-da6c62e22ca8",
|
145 |
+
"metadata": {},
|
146 |
+
"outputs": [],
|
147 |
+
"source": []
|
148 |
+
}
|
149 |
+
],
|
150 |
+
"metadata": {
|
151 |
+
"kernelspec": {
|
152 |
+
"display_name": "Python 3 (ipykernel)",
|
153 |
+
"language": "python",
|
154 |
+
"name": "python3"
|
155 |
+
},
|
156 |
+
"language_info": {
|
157 |
+
"codemirror_mode": {
|
158 |
+
"name": "ipython",
|
159 |
+
"version": 3
|
160 |
+
},
|
161 |
+
"file_extension": ".py",
|
162 |
+
"mimetype": "text/x-python",
|
163 |
+
"name": "python",
|
164 |
+
"nbconvert_exporter": "python",
|
165 |
+
"pygments_lexer": "ipython3",
|
166 |
+
"version": "3.9.11"
|
167 |
+
}
|
168 |
+
},
|
169 |
+
"nbformat": 4,
|
170 |
+
"nbformat_minor": 5
|
171 |
+
}
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
pandas
|
3 |
+
transformers
|