Plus: Addes the .csv from huawei

src/predict.py

@@ -1,5 +1,6 @@
 from pathlib import Path
 import torch
+import numpy as np
 
 from .tokenizer import load_tokenizer, preprocessing_text
 from .model import load_model
@@ -17,6 +18,24 @@ tokenizer = load_tokenizer(model_name)
 model = load_model(checkpoint_path, model_name, num_labels, divice)
 
 
+RETURN_VALUES =[
+    "target_sentiment_negative",
+    "target_sentiment_neutral",
+    "target_sentiment_positive",
+    "companies_sentiment_negative",
+    "companies_sentiment_neutral",
+    "companies_sentiment_positive",
+    "consumers_sentiment_negative",
+    "consumers_sentiment_neutral",
+    "consumers_sentiment_positive"
+]
+
+def filter(preds, threshold=0.5):
+    bool = preds > threshold
+    indices = np.where(bool)[0]
+    filtered_values = {RETURN_VALUES[index]: preds[index] for index in indices}
+    return filtered_values
+
 def get_predict(text):
     inputs = preprocessing_text(text, tokenizer)
     input_ids = inputs["input_ids"].to(divice)
@@ -24,4 +43,5 @@ def get_predict(text):
     token_type_ids = inputs["token_type_ids"].to(divice)
     outputs = model(input_ids, attention_mask, token_type_ids)
     preds = torch.sigmoid(outputs).detach().cpu().numpy()
+    preds = filter(preds[0])
     return preds