Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -2,7 +2,7 @@ import streamlit as st
|
|
2 |
import torch
|
3 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
4 |
import matplotlib.pyplot as plt
|
5 |
-
import numpy
|
6 |
|
7 |
@st.cache_resource
|
8 |
def load_model():
|
@@ -12,9 +12,11 @@ def load_model():
|
|
12 |
|
13 |
tokenizer, model = load_model()
|
14 |
|
15 |
-
st.title("Sentiment Analysis App")
|
16 |
|
17 |
text = st.text_input("Enter text to analyze:")
|
|
|
|
|
18 |
if st.button("Analyze") and text:
|
19 |
encoding = tokenizer.encode_plus(text, return_tensors="pt", padding=True, truncation=True)
|
20 |
input_ids = encoding["input_ids"]
|
@@ -23,19 +25,31 @@ if st.button("Analyze") and text:
|
|
23 |
with torch.no_grad():
|
24 |
output = model(input_ids, attention_mask)
|
25 |
logits = output.logits.squeeze()
|
26 |
-
|
27 |
-
# Determine the number of sentiment classes from the model output
|
28 |
num_classes = logits.shape[0]
|
29 |
sentiments = ["Very Negative", "Negative", "Neutral", "Positive", "Very Positive"][:num_classes]
|
30 |
|
|
|
|
|
|
|
31 |
prediction = int(torch.argmax(logits))
|
32 |
sentiment = sentiments[prediction]
|
33 |
-
st.write(f"Sentiment: {sentiment}")
|
34 |
|
35 |
-
|
|
|
36 |
|
37 |
fig, ax = plt.subplots()
|
38 |
-
|
39 |
-
ax.
|
40 |
-
|
41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import torch
|
3 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
4 |
import matplotlib.pyplot as plt
|
5 |
+
import numpy as np
|
6 |
|
7 |
@st.cache_resource
|
8 |
def load_model():
|
|
|
12 |
|
13 |
tokenizer, model = load_model()
|
14 |
|
15 |
+
st.title("Advanced Sentiment Analysis App")
|
16 |
|
17 |
text = st.text_input("Enter text to analyze:")
|
18 |
+
threshold = st.slider("Set sentiment strength threshold:", 0.0, 1.0, 0.5, 0.01)
|
19 |
+
|
20 |
if st.button("Analyze") and text:
|
21 |
encoding = tokenizer.encode_plus(text, return_tensors="pt", padding=True, truncation=True)
|
22 |
input_ids = encoding["input_ids"]
|
|
|
25 |
with torch.no_grad():
|
26 |
output = model(input_ids, attention_mask)
|
27 |
logits = output.logits.squeeze()
|
28 |
+
|
|
|
29 |
num_classes = logits.shape[0]
|
30 |
sentiments = ["Very Negative", "Negative", "Neutral", "Positive", "Very Positive"][:num_classes]
|
31 |
|
32 |
+
softmax = torch.nn.Softmax(dim=0)
|
33 |
+
probabilities = softmax(logits).numpy()
|
34 |
+
|
35 |
prediction = int(torch.argmax(logits))
|
36 |
sentiment = sentiments[prediction]
|
37 |
+
st.write(f"Detected Sentiment: {sentiment}")
|
38 |
|
39 |
+
# Normalize scores for display
|
40 |
+
values = probabilities.tolist()
|
41 |
|
42 |
fig, ax = plt.subplots()
|
43 |
+
colors = plt.cm.coolwarm(np.linspace(0, 1, num_classes))
|
44 |
+
bars = ax.bar(sentiments, values, color=colors)
|
45 |
+
|
46 |
+
# Highlight bars that pass the threshold
|
47 |
+
for bar, value in zip(bars, values):
|
48 |
+
if value > threshold:
|
49 |
+
bar.set_alpha(1.0) # Solid color for high confidence
|
50 |
+
else:
|
51 |
+
bar.set_alpha(0.5) # Faded color for low confidence
|
52 |
+
|
53 |
+
ax.set_title("Sentiment Analysis Scores with Confidence Threshold")
|
54 |
+
ax.set_ylabel("Confidence")
|
55 |
+
st.pyplot(fig)
|