Upload folder using huggingface_hub

README.md

@@ -1,3 +1,112 @@
 ---
-license: apache-2.0
+license: mit
+tags:
+  - pytorch
+  - anomaly-detection
+  - health
+  - fitness
+  - overtraining
+  - autoencoder
+  - whoop
+library_name: pytorch
 ---
+
+# WHOOP Overtraining Detection Model (CONV)
+
+This is a CONV autoencoder model trained to detect overtraining patterns in WHOOP fitness data.
+
+## Model Description
+
+This model analyzes physiological metrics (HRV, resting heart rate, sleep quality, strain, etc.) to identify
+unusual patterns that may indicate overtraining, illness, or injury risk.
+
+**How it works:**
+1. The autoencoder learns to reconstruct "normal" physiological patterns
+2. High reconstruction error indicates the pattern is unusual/anomalous
+3. Anomaly scores are computed as standard deviations from normal
+
+## Model Architecture
+
+- **Type**: CONV Autoencoder
+- **Input**: 7-day sequences of 33 physiological features
+- **Latent Size**: 32
+- **Parameters**: N/A
+
+## Performance Metrics
+
+| Metric | Value |
+|--------|-------|
+| Test Loss | 0.2008 |
+| Error Mean | 0.1413 |
+| Error Std | 0.0378 |
+| Anomaly Threshold | 0.2113 |
+
+## Usage
+
+```python
+import torch
+import joblib
+from huggingface_hub import hf_hub_download
+
+# Download model and scaler
+model_path = hf_hub_download(repo_id="iserrano0511/whoop-overtraining-detector", filename="model.pt")
+scaler_path = hf_hub_download(repo_id="iserrano0511/whoop-overtraining-detector", filename="scaler.pkl")
+
+# Load checkpoint
+checkpoint = torch.load(model_path, map_location='cpu', weights_only=False)
+scaler = joblib.load(scaler_path)
+
+# Create model (you'll need the model.py from the repo)
+from model import create_autoencoder
+
+model = create_autoencoder(
+    model_type=checkpoint['model_type'],
+    input_size=checkpoint['input_size'],
+    seq_len=checkpoint['seq_len'],
+    hidden_size=checkpoint['hidden_size'],
+    latent_size=checkpoint['latent_size'],
+    num_layers=checkpoint['num_layers'],
+    dropout=checkpoint['dropout']
+)
+model.load_state_dict(checkpoint['model_state_dict'])
+model.eval()
+
+# Use for inference
+# data should be shape (batch, 7, 33) - 7 days, 33 features
+# scaled using the provided scaler
+with torch.no_grad():
+    x_recon, latent = model(data)
+    error = ((data - x_recon) ** 2).mean(dim=(1, 2))  # Per-sample MSE
+
+# Check for anomalies
+threshold = checkpoint['threshold']
+is_anomaly = error > threshold
+```
+
+## Features Used
+
+The model uses 33 engineered features including:
+- **Physiological**: HRV, resting heart rate, respiratory rate, skin temp deviation
+- **Sleep**: Hours, efficiency, deep/REM ratios, sleep debt indicator
+- **Strain**: Day strain, cumulative strain (3d/7d), strain-recovery ratio
+- **Trends**: Rolling 3-day and 7-day averages, day-over-day changes
+
+## Limitations
+
+- Trained on synthetic WHOOP-style data
+- Best used as a screening tool, not diagnostic
+- Individual baselines vary - results should be interpreted in context
+
+## Citation
+
+If you use this model, please cite:
+
+```
+@misc{whoop-overtraining-detector,
+  author = {Your Name},
+  title = {WHOOP Overtraining Detection Model},
+  year = {2024},
+  publisher = {Hugging Face},
+  url = {https://huggingface.co/iserrano0511/whoop-overtraining-detector}
+}
+```

config.json

@@ -0,0 +1,12 @@
+{
+  "model_type": "conv",
+  "input_size": 33,
+  "seq_len": 7,
+  "hidden_size": 64,
+  "latent_size": 32,
+  "num_layers": 2,
+  "dropout": 0.2,
+  "threshold": 0.21133320033550262,
+  "train_error_mean": 0.1413002759218216,
+  "train_error_std": 0.03776929900050163
+}

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f299b4a220670dd420e6e4621be2a659c87b8631bfbd23b2cc93d5ce88378448
+size 159887

scaler.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e841dacdf6c68b76e8f90e3effae25577647fb293d6bb24e336de5a99f6bbf4
+size 927