Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -27,15 +27,47 @@ class RaceClassifier(nn.Module):
|
|
27 |
return self.out(output)
|
28 |
|
29 |
|
30 |
-
|
31 |
0: "African American",
|
32 |
1: "Asian",
|
33 |
2: "Latin",
|
34 |
3: "White"
|
35 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
model_race = RaceClassifier(n_classes=4)
|
37 |
model_race.to(device)
|
38 |
-
model_race.load_state_dict(torch.load('best_model_race.pt'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
40 |
def predict(*text):
|
41 |
tweets = [tweet for tweet in text if tweet]
|
@@ -55,22 +87,16 @@ def predict(*text):
|
|
55 |
input_ids = encoded_sentences["input_ids"].to(device)
|
56 |
attention_mask = encoded_sentences["attention_mask"].to(device)
|
57 |
|
58 |
-
model_race
|
59 |
-
|
60 |
-
outputs = model_race(input_ids, attention_mask)
|
61 |
-
probs = torch.nn.functional.softmax(outputs, dim=1)
|
62 |
-
predictions = torch.argmax(outputs, dim=1)
|
63 |
-
predictions = predictions.cpu().numpy()
|
64 |
-
|
65 |
-
output_string = "RACE\n Probabilities:\n"
|
66 |
-
for i, prob in enumerate(probs[0]):
|
67 |
-
print(f"{labels[i]} = {round(prob.item() * 100, 2)}%")
|
68 |
-
output_string += f"{labels[i]} = {round(prob.item() * 100, 2)}%\n"
|
69 |
|
70 |
-
|
71 |
-
|
|
|
|
|
|
|
72 |
|
73 |
-
return
|
74 |
|
75 |
|
76 |
max_textboxes = 20
|
|
|
27 |
return self.out(output)
|
28 |
|
29 |
|
30 |
+
race_labels = {
|
31 |
0: "African American",
|
32 |
1: "Asian",
|
33 |
2: "Latin",
|
34 |
3: "White"
|
35 |
}
|
36 |
+
|
37 |
+
orientation_labels = {
|
38 |
+
0: "Heterosexual",
|
39 |
+
1: "LGBTQ"
|
40 |
+
}
|
41 |
+
|
42 |
model_race = RaceClassifier(n_classes=4)
|
43 |
model_race.to(device)
|
44 |
+
model_race.load_state_dict(torch.load('best_model_race.pt'))
|
45 |
+
|
46 |
+
model_orientation = RaceClassifier(n_classes=2)
|
47 |
+
model_orientation.to(device)
|
48 |
+
model_orientation.load_state_dict(torch.load('best_model_orientation_last.pt'))
|
49 |
+
|
50 |
+
|
51 |
+
def evaluate(model, input, mask):
|
52 |
+
model.eval()
|
53 |
+
with torch.no_grad():
|
54 |
+
outputs = model(input, mask)
|
55 |
+
probs = torch.nn.functional.softmax(outputs, dim=1)
|
56 |
+
predictions = torch.argmax(outputs, dim=1)
|
57 |
+
predictions = predictions.cpu().numpy()
|
58 |
+
return probs, predictions
|
59 |
+
|
60 |
+
|
61 |
+
def write_output(probs, predictions, title, labels):
|
62 |
+
output_string = f"{title.upper()}\n Probabilities:\n"
|
63 |
+
for i, prob in enumerate(probs[0]):
|
64 |
+
print(f"{labels[i]} = {round(prob.item() * 100, 2)}%")
|
65 |
+
output_string += f"{labels[i]} = {round(prob.item() * 100, 2)}%\n"
|
66 |
+
|
67 |
+
output_string += f"Predicted as: {labels[predictions[0]]}\n"
|
68 |
+
|
69 |
+
return output_string
|
70 |
+
|
71 |
|
72 |
def predict(*text):
|
73 |
tweets = [tweet for tweet in text if tweet]
|
|
|
87 |
input_ids = encoded_sentences["input_ids"].to(device)
|
88 |
attention_mask = encoded_sentences["attention_mask"].to(device)
|
89 |
|
90 |
+
race_probs, race_predictions = evaluate(model_race, input_ids, attention_mask)
|
91 |
+
orientation_probs, orientation_predictions = evaluate(model_orientation, input_ids, attention_mask)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
+
final_output = str()
|
94 |
+
final_output += write_output(race_probs, race_predictions, "race", race_labels)
|
95 |
+
final_output += "\n"
|
96 |
+
final_output += write_output(orientation_probs, orientation_predictions, "sexual orientation", orientation_labels)
|
97 |
+
final_output += "\n"
|
98 |
|
99 |
+
return final_output
|
100 |
|
101 |
|
102 |
max_textboxes = 20
|