iserrano0511/whoop-overtraining-detector

WHOOP Overtraining Detection

A convolutional autoencoder for detecting overtraining patterns in WHOOP fitness data.

Quick Start

Installation

pip install torch pandas numpy scikit-learn joblib huggingface_hub

Usage

Option 1: Using inference.py (Recommended)

Download inference.py from this repo, then:

from inference import OvertainingPredictor

# Load model
predictor = OvertainingPredictor.from_pretrained("iserrano0511/whoop-overtraining-detector")

# Predict from CSV
results = predictor.predict_from_csv("your_whoop_data.csv")

# Print summary
print(predictor.summary(results))

# Check anomalies
for r in results:
    if r.is_anomaly:
        print(f"{r.date}: {r.risk_level} risk (score: {r.anomaly_score:.2f})")

Option 2: Manual Loading

import torch
import joblib
from huggingface_hub import hf_hub_download

# Download files
model_path = hf_hub_download("iserrano0511/whoop-overtraining-detector", "model.pt")
scaler_path = hf_hub_download("iserrano0511/whoop-overtraining-detector", "scaler.pkl")
config_path = hf_hub_download("iserrano0511/whoop-overtraining-detector", "config.json")

# Load
checkpoint = torch.load(model_path, map_location='cpu', weights_only=False)
scaler = joblib.load(scaler_path)

Model Details

Property	Value
Architecture	CONV Autoencoder
Input Shape	(batch, 7, 33)
Latent Size	32
Anomaly Threshold	0.2113

How It Works

1. Input: 7 consecutive days of physiological data (33 features)
2. Processing: Autoencoder reconstructs the input
3. Detection: High reconstruction error = anomaly
4. Output: Risk level (normal/elevated/high/critical)

Risk Levels

Level	Score	Meaning
Normal	< 1.5 std	Typical patterns
Elevated	1.5-2.5 std	Minor deviation
High	2.5-3.5 std	Significant anomaly
Critical	> 3.5 std	Major deviation

Required Data Columns

Your CSV needs these columns:

• user_id, date
• hrv, resting_heart_rate, hrv_baseline, rhr_baseline
• respiratory_rate, skin_temp_deviation
• sleep_hours, sleep_efficiency, deep_sleep_hours, rem_sleep_hours
• wake_ups, time_to_fall_asleep_min
• day_strain, recovery_score

Files

• model.pt - PyTorch model weights
• scaler.pkl - Data scaler (joblib)
• config.json - Model configuration
• inference.py - Self-contained inference module

Limitations

• Trained on synthetic WHOOP-style data
• Screening tool only, not diagnostic
• Requires 7+ days of consecutive data

Citation

@misc{whoop-overtraining,
  author = {iserrano0511},
  title = {WHOOP Overtraining Detection},
  year = {2024},
  publisher = {Hugging Face},
  url = {https://huggingface.co/iserrano0511/whoop-overtraining-detector}
}