Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -3,16 +3,22 @@ import torch
|
|
3 |
from datasets import load_dataset
|
4 |
from console_logging.console import Console
|
5 |
import numpy as np
|
|
|
|
|
|
|
|
|
|
|
6 |
console = Console()
|
7 |
|
8 |
dataset = load_dataset("zeroshot/twitter-financial-news-sentiment", )
|
9 |
|
10 |
-
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
11 |
|
12 |
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
|
13 |
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
14 |
|
15 |
-
labels = [label for label in dataset['train'].features.keys() if label not in ['text']]
|
|
|
|
|
16 |
|
17 |
def preprocess_data(examples):
|
18 |
# take a batch of texts
|
@@ -20,7 +26,22 @@ def preprocess_data(examples):
|
|
20 |
# encode them
|
21 |
encoding = tokenizer(text, padding="max_length", truncation=True, max_length=128)
|
22 |
# add labels
|
23 |
-
labels_batch = {k: examples[k] for k in examples.keys() if k in labels}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
# create numpy array of shape (batch_size, num_labels)
|
25 |
labels_matrix = np.zeros((len(text), len(labels)))
|
26 |
# fill numpy array
|
@@ -38,7 +59,8 @@ encoded_dataset.set_format("torch")
|
|
38 |
id2label = {idx:label for idx, label in enumerate(labels)}
|
39 |
label2id = {label:idx for idx, label in enumerate(labels)}
|
40 |
|
41 |
-
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased",
|
|
|
42 |
num_labels=len(labels),
|
43 |
id2label=id2label,
|
44 |
label2id=label2id)
|
@@ -46,8 +68,6 @@ model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased",
|
|
46 |
batch_size = 8
|
47 |
metric_name = "f1"
|
48 |
|
49 |
-
from transformers import TrainingArguments, Trainer
|
50 |
-
|
51 |
args = TrainingArguments(
|
52 |
f"bert-finetuned-sem_eval-english",
|
53 |
evaluation_strategy = "epoch",
|
@@ -62,11 +82,6 @@ args = TrainingArguments(
|
|
62 |
#push_to_hub=True,
|
63 |
)
|
64 |
|
65 |
-
|
66 |
-
from sklearn.metrics import f1_score, roc_auc_score, accuracy_score
|
67 |
-
from transformers import EvalPrediction
|
68 |
-
import torch
|
69 |
-
|
70 |
# source: https://jesusleal.io/2021/04/21/Longformer-multilabel-classification/
|
71 |
def multi_label_metrics(predictions, labels, threshold=0.5):
|
72 |
# first, apply sigmoid on predictions which are of shape (batch_size, num_labels)
|
@@ -106,22 +121,4 @@ trainer = Trainer(
|
|
106 |
|
107 |
trainer.train()
|
108 |
|
109 |
-
trainer.evaluate()
|
110 |
-
|
111 |
-
"""
|
112 |
-
|
113 |
-
categories = ('Car in good condition','Damaged Car')
|
114 |
-
|
115 |
-
def is_car(x) : return x[0].isupper()
|
116 |
-
|
117 |
-
def image_classifier(img):
|
118 |
-
pred,index,probs = learn.predict(img)
|
119 |
-
return dict(zip(categories, map(float,probs)))
|
120 |
-
|
121 |
-
# image = gr.inputs.Image(shape=(192,192))
|
122 |
-
image = gr.components.Image(shape=(192,192))
|
123 |
-
label = gr.components.Label()
|
124 |
-
examples = ['./car.jpg','./crash.jpg','./carf.jpg']
|
125 |
-
|
126 |
-
intf = gr.Interface(fn= image_classifier,inputs=image,outputs=label,examples=examples)
|
127 |
-
intf.launch()"""
|
|
|
3 |
from datasets import load_dataset
|
4 |
from console_logging.console import Console
|
5 |
import numpy as np
|
6 |
+
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
7 |
+
from transformers import TrainingArguments, Trainer
|
8 |
+
from sklearn.metrics import f1_score, roc_auc_score, accuracy_score
|
9 |
+
from transformers import EvalPrediction
|
10 |
+
import torch
|
11 |
console = Console()
|
12 |
|
13 |
dataset = load_dataset("zeroshot/twitter-financial-news-sentiment", )
|
14 |
|
|
|
15 |
|
16 |
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
|
17 |
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
18 |
|
19 |
+
#labels = [label for label in dataset['train'].features.keys() if label not in ['text']]
|
20 |
+
|
21 |
+
labels = ["Bearish", "Bullish", "Neutral"]
|
22 |
|
23 |
def preprocess_data(examples):
|
24 |
# take a batch of texts
|
|
|
26 |
# encode them
|
27 |
encoding = tokenizer(text, padding="max_length", truncation=True, max_length=128)
|
28 |
# add labels
|
29 |
+
#labels_batch = {k: examples[k] for k in examples.keys() if k in labels}
|
30 |
+
labels_batch = {'Bearish': [], 'Bullish': [], 'Neutral': []}
|
31 |
+
for i in range (len(examples['label'])):
|
32 |
+
labels_batch["Bearish"].append(False)
|
33 |
+
labels_batch["Bullish"].append(False)
|
34 |
+
labels_batch["Neutral"].append(False)
|
35 |
+
|
36 |
+
if examples['label'][i] == 0:
|
37 |
+
labels_batch["Bearish"][i] = True
|
38 |
+
|
39 |
+
elif examples['label'][i] == 1:
|
40 |
+
labels_batch["Bullish"][i] = True
|
41 |
+
|
42 |
+
else:
|
43 |
+
labels_batch["Neutral"][i] = True
|
44 |
+
|
45 |
# create numpy array of shape (batch_size, num_labels)
|
46 |
labels_matrix = np.zeros((len(text), len(labels)))
|
47 |
# fill numpy array
|
|
|
59 |
id2label = {idx:label for idx, label in enumerate(labels)}
|
60 |
label2id = {label:idx for idx, label in enumerate(labels)}
|
61 |
|
62 |
+
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased",
|
63 |
+
problem_type="multi_label_classification",
|
64 |
num_labels=len(labels),
|
65 |
id2label=id2label,
|
66 |
label2id=label2id)
|
|
|
68 |
batch_size = 8
|
69 |
metric_name = "f1"
|
70 |
|
|
|
|
|
71 |
args = TrainingArguments(
|
72 |
f"bert-finetuned-sem_eval-english",
|
73 |
evaluation_strategy = "epoch",
|
|
|
82 |
#push_to_hub=True,
|
83 |
)
|
84 |
|
|
|
|
|
|
|
|
|
|
|
85 |
# source: https://jesusleal.io/2021/04/21/Longformer-multilabel-classification/
|
86 |
def multi_label_metrics(predictions, labels, threshold=0.5):
|
87 |
# first, apply sigmoid on predictions which are of shape (batch_size, num_labels)
|
|
|
121 |
|
122 |
trainer.train()
|
123 |
|
124 |
+
trainer.evaluate()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|