# Quickstart

Load a FELA-PdM head and run one real sensor window on CPU. Uses the self contained loader
in `../modeling.py`.

## Steps

1. Install the pinned requirements (CPU PyTorch):

   ```bash
   pip install -r requirements.txt
   ```

2. Get the weights. One safetensors file per head ships in this repo
   (`cmapss_FD001.safetensors`, `cwru.safetensors`) beside `config.json`. Point at the repo
   directory and pass `--variant` to pick the head:

   ```bash
   export FELA_PDM_WEIGHTS=/path/to/weights_dir   # holds <variant>.safetensors + config.json
   ```

3. Run:

   ```bash
   python run.py --variant cmapss_FD001   # remaining useful life
   python run.py --variant cwru           # bearing fault class
   ```

## Few line load from Python

```python
from modeling import load_model
m = load_model("/path/to/weights_dir", variant="cmapss_FD001")  # dir, .pt, or HF repo id
rul = m.predict(window)   # window: (1, 30, 14) sensor cycles
```

The C-MAPSS RUL head expects 30 cycles of 14 sensors (min max normalized on the training
statistics). The CWRU head expects 2048 raw vibration samples (12 kHz, per signal
standardized). See `modeling.preprocess_cmapss` and `modeling.preprocess_cwru` for the exact
preprocessing, and `modeling.validate_window` for input validation.