# LWMTemporal Examples

This directory contains example scripts demonstrating how to use the LWMTemporal package.

## Quick Start Examples

### 1. Masked Reconstruction (`example_reconstruction.py`)

Demonstrates how to:
- Load wireless channel data
- Tokenize complex channels
- Mask random positions
- Reconstruct using the pretrained model

```bash
python examples/example_reconstruction.py
```

### 2. Channel Prediction Inference (`inference_channel_prediction.py`)

Run inference with a fine-tuned channel prediction model:

```bash
python examples/inference_channel_prediction.py
```

Expected output: Per-step NMSE around -20 dB

### 3. Train Channel Prediction (`train_channel_prediction.py`)

Fine-tune the model for channel prediction:

```bash
python examples/train_channel_prediction.py
```

This will:
- Load pretrained weights
- Fine-tune on your dataset
- Save checkpoints to `models/`
- Generate visualizations in `figs/predictions/`

## Using the CLI

The package also provides command-line interfaces:

### Channel Prediction

```bash
python -m LWMTemporal.cli.channel_prediction \
  --data_path examples/data/city_8_tempe_3p5_20_32_32.p \
  --pretrained checkpoints/m18_cp.pth \
  --inference_only \
  --val_limit 100 \
  --device cpu
```

### Pretraining

```bash
python -m LWMTemporal.cli.pretrain \
  --data_dir examples/data/ \
  --save_prefix models/pretrained \
  --epochs 100 \
  --batch_size 32 \
  --device cuda
```

## Data Format

Example data files are in `examples/data/`. See `examples/data/README.md` for details on the expected format.

## Checkpoints

Pretrained checkpoints are in `checkpoints/`. See `checkpoints/README.md` for available models and loading instructions.