ClassifierNeuralNet

A simple neural network text classifier that converts English text into one of 100 predefined actions using Sentence Transformers and PyTorch.

Features

• Uses all-MiniLM-L6-v2 Sentence Transformer embeddings
• Multi-layer neural network (MLP) classifier
• Supports up to 100 predefined action classes
• Simple training loop with PyTorch
• Easy inference on new text

---

Architecture

English Text
      │
      ▼
SentenceTransformer
(all-MiniLM-L6-v2)
      │
384-dimensional embedding
      │
      ▼
Linear(384 → 256)
      │
ReLU
      │
Dropout(0.2)
      │
Linear(256 → 128)
      │
ReLU
      │
Linear(128 → 100)
      │
      ▼
Predicted Action ID

---

Training Data

Replace the example dataset with your own.

texts = [
    "turn on lights",
    "play music",
    "book cab",
    "switch off fan",
    "order food"
]

labels = [
    5,
    12,
    37,
    5,
    22
]

Each label is an integer representing one predefined action.

---

Train

Run:

python classifier.py

Example output:

Epoch 1, Loss: 2.94
Epoch 2, Loss: 2.10
...
Training complete

---

Inference

pred, probs = predict("play some songs")

print(pred)

Output:

12

---

Model

• Embedding Model: all-MiniLM-L6-v2
• Embedding Size: 384
• Hidden Layers:
  • 256 neurons
  • 128 neurons
• Output Classes: 100
• Loss: CrossEntropyLoss
• Optimizer: Adam

---

Requirements

• Python 3.9+
• PyTorch
• sentence-transformers

---

License

MIT License