Update app.py
Browse files
app.py
CHANGED
@@ -359,6 +359,7 @@ demo = gr.Interface(
|
|
359 |
|
360 |
demo.launch()
|
361 |
'''
|
|
|
362 |
import gradio as gr
|
363 |
from transformers import TFBertForSequenceClassification, BertTokenizer
|
364 |
import tensorflow as tf
|
@@ -533,7 +534,122 @@ demo = gr.TabbedInterface(
|
|
533 |
)
|
534 |
|
535 |
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
536 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
537 |
|
538 |
|
539 |
|
|
|
359 |
|
360 |
demo.launch()
|
361 |
'''
|
362 |
+
'''
|
363 |
import gradio as gr
|
364 |
from transformers import TFBertForSequenceClassification, BertTokenizer
|
365 |
import tensorflow as tf
|
|
|
534 |
)
|
535 |
|
536 |
demo.launch()
|
537 |
+
'''
|
538 |
+
import gradio as gr
|
539 |
+
from transformers import TFBertForSequenceClassification, BertTokenizer
|
540 |
+
import tensorflow as tf
|
541 |
+
from sklearn.metrics import accuracy_score, f1_score, classification_report
|
542 |
+
import numpy as np
|
543 |
|
544 |
+
# Load models
|
545 |
+
model = TFBertForSequenceClassification.from_pretrained("shrish191/sentiment-bert")
|
546 |
+
tokenizer = BertTokenizer.from_pretrained("shrish191/sentiment-bert")
|
547 |
+
LABELS = {0: "Neutral", 1: "Positive", 2: "Negative"}
|
548 |
+
|
549 |
+
# Load fallback model
|
550 |
+
fallback_model_name = "cardiffnlp/twitter-roberta-base-sentiment"
|
551 |
+
fallback_tokenizer = AutoTokenizer.from_pretrained(fallback_model_name)
|
552 |
+
fallback_model = AutoModelForSequenceClassification.from_pretrained(fallback_model_name)
|
553 |
+
|
554 |
+
def analyze_text(text, true_label=None):
|
555 |
+
try:
|
556 |
+
# Main model prediction
|
557 |
+
inputs = tokenizer(text, return_tensors="tf", truncation=True, padding=True)
|
558 |
+
outputs = model(inputs)
|
559 |
+
probs = tf.nn.softmax(outputs.logits, axis=1)
|
560 |
+
main_pred = LABELS[tf.argmax(probs, axis=1).numpy()[0]]
|
561 |
+
|
562 |
+
# Fallback model prediction
|
563 |
+
fallback_inputs = fallback_tokenizer(text, return_tensors='pt', truncation=True, padding=True)
|
564 |
+
with torch.no_grad():
|
565 |
+
fallback_outputs = fallback_model(**fallback_inputs)
|
566 |
+
fallback_scores = softmax(fallback_outputs.logits.numpy()[0])
|
567 |
+
fallback_pred = ['Negative', 'Neutral', 'Positive'][np.argmax(fallback_scores)]
|
568 |
+
|
569 |
+
# Initialize results
|
570 |
+
result = f"""Main Model Prediction: {main_pred}
|
571 |
+
Fallback Model Prediction: {fallback_pred}"""
|
572 |
+
|
573 |
+
# Calculate metrics if true label provided
|
574 |
+
if true_label:
|
575 |
+
# Convert labels to numerical format
|
576 |
+
label_map = {v: k for k, v in LABELS.items()}
|
577 |
+
y_true = [label_map[true_label]]
|
578 |
+
|
579 |
+
# Main model metrics
|
580 |
+
y_pred_main = [label_map[main_pred]]
|
581 |
+
main_acc = accuracy_score(y_true, y_pred_main)
|
582 |
+
main_f1 = f1_score(y_true, y_pred_main, average='weighted')
|
583 |
+
|
584 |
+
# Fallback model metrics
|
585 |
+
fallback_label_map = {'Negative': 2, 'Neutral': 0, 'Positive': 1}
|
586 |
+
y_pred_fallback = [fallback_label_map[fallback_pred]]
|
587 |
+
fallback_acc = accuracy_score(y_true, y_pred_fallback)
|
588 |
+
fallback_f1 = f1_score(y_true, y_pred_fallback, average='weighted')
|
589 |
+
|
590 |
+
# Classification report
|
591 |
+
report = classification_report(
|
592 |
+
y_true, y_pred_main,
|
593 |
+
target_names=LABELS.values(),
|
594 |
+
output_dict=True
|
595 |
+
)
|
596 |
+
|
597 |
+
# Format metrics
|
598 |
+
metrics = f"""
|
599 |
+
\n\nPERFORMANCE METRICS (Single Sample):
|
600 |
+
------------------------------------
|
601 |
+
Main Model:
|
602 |
+
Accuracy: {main_acc:.4f}
|
603 |
+
F1 Score: {main_f1:.4f}
|
604 |
+
|
605 |
+
Fallback Model:
|
606 |
+
Accuracy: {fallback_acc:.4f}
|
607 |
+
F1 Score: {fallback_f1:.4f}
|
608 |
+
|
609 |
+
Classification Report:
|
610 |
+
{classification_report(y_true, y_pred_main, target_names=LABELS.values())}
|
611 |
+
"""
|
612 |
+
|
613 |
+
result += metrics
|
614 |
+
|
615 |
+
return result
|
616 |
+
|
617 |
+
except Exception as e:
|
618 |
+
return f"Error: {str(e)}"
|
619 |
+
|
620 |
+
# Gradio interface
|
621 |
+
demo = gr.Interface(
|
622 |
+
fn=analyze_text,
|
623 |
+
inputs=[
|
624 |
+
gr.Textbox(
|
625 |
+
label="Input Text",
|
626 |
+
placeholder="Enter text to analyze...",
|
627 |
+
lines=4
|
628 |
+
),
|
629 |
+
gr.Dropdown(
|
630 |
+
label="True Label (optional, for metrics)",
|
631 |
+
choices=list(LABELS.values()),
|
632 |
+
value=None
|
633 |
+
)
|
634 |
+
],
|
635 |
+
outputs=gr.Textbox(
|
636 |
+
label="Analysis Results",
|
637 |
+
lines=10
|
638 |
+
),
|
639 |
+
title="Sentiment Analysis with Performance Metrics",
|
640 |
+
description="""Enter text and optionally select true label to generate:
|
641 |
+
- Predictions from both models
|
642 |
+
- Accuracy scores
|
643 |
+
- F1 scores
|
644 |
+
- Classification report""",
|
645 |
+
examples=[
|
646 |
+
["I absolutely love this new feature!", "Positive"],
|
647 |
+
["This is the worst experience ever!", "Negative"],
|
648 |
+
["The product seems okay, nothing special.", "Neutral"]
|
649 |
+
]
|
650 |
+
)
|
651 |
+
|
652 |
+
demo.launch()
|
653 |
|
654 |
|
655 |
|