Spaces:
Runtime error
Runtime error
thotran
commited on
Commit
•
9993f32
1
Parent(s):
952a624
reduced wait time
Browse files- .DS_Store +0 -0
- app.py +53 -37
- data/sub.csv +0 -0
- requirements.txt +2 -2
.DS_Store
CHANGED
Binary files a/.DS_Store and b/.DS_Store differ
|
|
app.py
CHANGED
@@ -7,13 +7,15 @@ import torch.nn.functional as F
|
|
7 |
from torch.utils.data import TensorDataset, DataLoader, Dataset
|
8 |
from sklearn.metrics import roc_auc_score
|
9 |
import re
|
10 |
-
from
|
11 |
from typing import *
|
12 |
import string
|
13 |
from sklearn.model_selection import train_test_split
|
14 |
from transformers import DistilBertTokenizer, AdamW
|
15 |
from transformers import DistilBertModel, DistilBertConfig, DistilBertForSequenceClassification
|
|
|
16 |
import streamlit as st
|
|
|
17 |
st.write("Please be patient model training takes 20+ mins :P")
|
18 |
#config constants
|
19 |
SEED = 42
|
@@ -58,7 +60,7 @@ tokenizer = DistilBertTokenizer.from_pretrained(PRE_TRAINED_MODEL_NAME)
|
|
58 |
|
59 |
token_lens = []
|
60 |
|
61 |
-
for txt in
|
62 |
tokens = tokenizer.encode(txt, max_length=512)
|
63 |
token_lens.append(len(tokens))
|
64 |
|
@@ -134,7 +136,7 @@ def train_epoch_for_hf(model, data_loader: DataLoader, device: torch.device, opt
|
|
134 |
"""
|
135 |
model.train()
|
136 |
|
137 |
-
for batch in
|
138 |
input_ids = batch["input_ids"].to(device)
|
139 |
attention_mask = batch["attention_mask"].to(device)
|
140 |
targets = batch["targets"].float().to(device)
|
@@ -152,7 +154,7 @@ def evaluate_for_hf(model, data_loader: DataLoader, device: torch.device):
|
|
152 |
losses = []
|
153 |
score = None
|
154 |
|
155 |
-
for idx, batch in enumerate(
|
156 |
input_ids = batch["input_ids"].to(device)
|
157 |
attention_mask = batch["attention_mask"].to(device)
|
158 |
targets = batch["targets"].float().to(device)
|
@@ -169,38 +171,52 @@ optimizer = AdamW(model.parameters(), lr=2e-5)
|
|
169 |
best_val_loss = 9999.
|
170 |
print('====START TRAINING====')
|
171 |
#training here
|
172 |
-
for epoch in
|
173 |
-
print('-' * 10)
|
174 |
-
train_epoch_for_hf(model=model, data_loader=train_dataloader, optimizer=optimizer, device=device)
|
175 |
-
_, tr_loss = evaluate_for_hf(model=model, data_loader=train_dataloader, device=device)
|
176 |
-
val_pred, val_loss = evaluate_for_hf(model=model, data_loader=val_dataloader, device=device)
|
177 |
-
y_pred_np = val_pred.numpy()
|
178 |
-
val_auc = roc_auc_score(df_val[labels].to_numpy(), y_pred_np)
|
179 |
-
if val_loss < best_val_loss:
|
180 |
-
best_val_loss = val_loss
|
181 |
#torch.save(model.state_dict(), 'distill_bert.pt')
|
182 |
-
print(f'Epoch {epoch + 1}/{EPOCHS}', f'train loss: {tr_loss:.4},', f'val loss: {val_loss:.4},', f'val auc: {val_auc:.4}')
|
183 |
# once model is saved and generated no need to re run :)
|
184 |
-
|
185 |
-
#
|
186 |
-
#
|
187 |
-
#
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
from torch.utils.data import TensorDataset, DataLoader, Dataset
|
8 |
from sklearn.metrics import roc_auc_score
|
9 |
import re
|
10 |
+
from stqdm import stqdm
|
11 |
from typing import *
|
12 |
import string
|
13 |
from sklearn.model_selection import train_test_split
|
14 |
from transformers import DistilBertTokenizer, AdamW
|
15 |
from transformers import DistilBertModel, DistilBertConfig, DistilBertForSequenceClassification
|
16 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
17 |
import streamlit as st
|
18 |
+
|
19 |
st.write("Please be patient model training takes 20+ mins :P")
|
20 |
#config constants
|
21 |
SEED = 42
|
|
|
60 |
|
61 |
token_lens = []
|
62 |
|
63 |
+
for txt in stqdm(data.comment_text,desc="tokenizing"):
|
64 |
tokens = tokenizer.encode(txt, max_length=512)
|
65 |
token_lens.append(len(tokens))
|
66 |
|
|
|
136 |
"""
|
137 |
model.train()
|
138 |
|
139 |
+
for batch in stqdm(data_loader, desc="training"):
|
140 |
input_ids = batch["input_ids"].to(device)
|
141 |
attention_mask = batch["attention_mask"].to(device)
|
142 |
targets = batch["targets"].float().to(device)
|
|
|
154 |
losses = []
|
155 |
score = None
|
156 |
|
157 |
+
for idx, batch in enumerate(stqdm(data_loader,desc="evaluating")):
|
158 |
input_ids = batch["input_ids"].to(device)
|
159 |
attention_mask = batch["attention_mask"].to(device)
|
160 |
targets = batch["targets"].float().to(device)
|
|
|
171 |
best_val_loss = 9999.
|
172 |
print('====START TRAINING====')
|
173 |
#training here
|
174 |
+
#for epoch in stqdm(range(EPOCHS)):
|
175 |
+
# print('-' * 10)
|
176 |
+
# train_epoch_for_hf(model=model, data_loader=train_dataloader, optimizer=optimizer, device=device)
|
177 |
+
# _, tr_loss = evaluate_for_hf(model=model, data_loader=train_dataloader, device=device)
|
178 |
+
# val_pred, val_loss = evaluate_for_hf(model=model, data_loader=val_dataloader, device=device)
|
179 |
+
# y_pred_np = val_pred.numpy()
|
180 |
+
# val_auc = roc_auc_score(df_val[labels].to_numpy(), y_pred_np)
|
181 |
+
# if val_loss < best_val_loss:
|
182 |
+
# best_val_loss = val_loss
|
183 |
#torch.save(model.state_dict(), 'distill_bert.pt')
|
184 |
+
# print(f'Epoch {epoch + 1}/{EPOCHS}', f'train loss: {tr_loss:.4},', f'val loss: {val_loss:.4},', f'val auc: {val_auc:.4}')
|
185 |
# once model is saved and generated no need to re run :)
|
186 |
+
|
187 |
+
#PUSH MODEL TO HF
|
188 |
+
#from huggingface_hub import notebook_login
|
189 |
+
#notebook_login()
|
190 |
+
#model.push_to_hub("tweetbert")
|
191 |
+
#tokenizer.push_to_hub("tweetbert")
|
192 |
+
|
193 |
+
#LOAD MODEL
|
194 |
+
model=model = AutoModelForSequenceClassification.from_pretrained("thotranexe/tweetbert")
|
195 |
+
model = model.to(device)
|
196 |
+
|
197 |
+
#TEST MODEL
|
198 |
+
#test_pred, test_loss = evaluate_for_hf(model=model, data_loader=test_dataloader, device=device)
|
199 |
+
#print('====TEST RESULT====')
|
200 |
+
#print(f'Log loss: {test_loss:.5}')
|
201 |
+
#y_pred_np = test_pred.numpy()
|
202 |
+
#test_auc = roc_auc_score(df_test[labels].to_numpy(), y_pred_np)
|
203 |
+
#print(f'ROC AUC: {test_auc:.5}')
|
204 |
+
|
205 |
+
#test_src_id = test.iloc[:, 0]
|
206 |
+
#test.drop(columns='id', inplace=True)
|
207 |
+
#test_labels.drop(columns='id', inplace=True)
|
208 |
+
#test_src = pd.concat((test, test_labels), axis=1)
|
209 |
+
|
210 |
+
#MAKE PREDICTIONS
|
211 |
+
#test_src_dataloader = create_data_loader(df=test_src, tokenizer=tokenizer, max_len=SEQ_SIZE, batch_size=1)
|
212 |
+
#prediction, _ = evaluate_for_hf(model=model, data_loader=test_src_dataloader, device=device)
|
213 |
+
#prediction = torch.sigmoid(prediction).numpy()
|
214 |
+
|
215 |
+
#SAVE RESULTS INTO SUBMISSION DATAFRAME
|
216 |
+
#sub[labels] = prediction
|
217 |
+
#sub.insert(1,"tweet",data.comment_text,True)
|
218 |
+
#sub.to_csv("sub.csv", encoding='utf-8', index=False)
|
219 |
+
#^commented above code, saved to csv to reduce wait/comput time
|
220 |
+
sub=pd.read_csv('./data/sub.csv',engine='python',encoding='utf-8', error_bad_lines=False)
|
221 |
+
sub.drop(index="id")
|
222 |
+
st.dataframe(sub)
|
data/sub.csv
ADDED
Binary file (71.3 MB). View file
|
|
requirements.txt
CHANGED
@@ -2,7 +2,7 @@ numpy
|
|
2 |
pandas
|
3 |
streamlit
|
4 |
torch
|
5 |
-
|
6 |
scikit-learn
|
7 |
transformers
|
8 |
-
ipywidgets
|
|
|
2 |
pandas
|
3 |
streamlit
|
4 |
torch
|
5 |
+
stqdm
|
6 |
scikit-learn
|
7 |
transformers
|
8 |
+
ipywidgets
|