Body Measurement Predictor
PyTorch regression model and Tkinter app forecasting physique changes after weight adjustments.
A machine learning playground that turned into a polished desktop tool: I built this pipeline to estimate how a client’s body measurements could shift after changing weight. From data prep to PyTorch training to a Tkinter GUI, it all lives here. Repo: github.com/zhaojinchu/BodyMeasurementPredictor.
Product flow
- Tkinter desktop app lets coaches plug in current stats, target weight, and instantly see projected measurement deltas.
- GUI constrains skeletal metrics (wrists, ankles, shoulder-to-crotch) so predictions stay realistic.
- Packaging checklist documents how to ship the tool with PyInstaller for offline use.
Model pipeline
-
preprocess.pycleans and augments datasets, simulating multiple weight-change scenarios before scaling features. - PyTorch regressors (
train2.py,train3.py) train on percentage deltas with early-stop checkpoints and scaler persistence. - Inference scripts reverse the scaling, clamp deltas, and output centimetre estimates that stay consistent across CLI and GUI.
Tooling & ops
- Saved scalers, model weights, and augmented CSVs live alongside code for reproducible retraining.
- Synthetic augmentation + StandardScaler combos cut down on overfitting and simplify iterative experiments.
- Troubleshooting docs cover data sanity checks and PyInstaller options for Windows environments.
Placeholder
Next up: add confidence intervals around each prediction and expand the dataset to support additional demographics.