Add demo prediction script for 1D CNN classifier

This script demonstrates the prediction process for a 1D CNN classifier using dummy tokenized sequences. It includes model initialization, forward pass, and output of probabilities.

Author: dmortalla

demo_predict.py

@@ -0,0 +1,41 @@
+"""
+Demo prediction script for the 1D CNN complaint classifier.
+
+This uses a dummy tokenized sequence to show the forward pass.
+Replace the dummy_ids with real token IDs from your tokenizer if desired.
+"""
+
+import torch
+from train_1d_cnn import ComplaintCNN  # ensure this matches your model class name
+
+
+def main():
+    # These hyperparameters must match how the model is defined in train_1d_cnn.py
+    vocab_size = 5000
+    embed_dim = 64
+    num_classes = 3
+    max_len = 50
+
+    model = ComplaintCNN(
+        vocab_size=vocab_size,
+        embed_dim=embed_dim,
+        num_classes=num_classes,
+    )
+
+    # Dummy batch of token ids: [batch_size, seq_len]
+    batch_size = 2
+    dummy_ids = torch.randint(0, vocab_size, (batch_size, max_len))
+
+    # Forward pass
+    logits = model(dummy_ids)
+    probs = torch.softmax(logits, dim=-1)
+
+    print(f"Input shape:   {dummy_ids.shape}")
+    print(f"Logits shape:  {logits.shape}")
+    print("Probabilities:")
+    print(probs)
+
+
+if __name__ == "__main__":
+    main()
+