How manh20060961815271-arch/matrix-neural works
This repository implements a cyclone-track forecasting workflow: it loads IBTrACS storm tracks, prepares relative-coordinate windows, trains a neural network to predict a point four steps ahead, then evaluates and plots predictions in absolute coordinates. The README provides no detail beyond the project name; the workflow and links below are based on the sampled source, so unsampled output-writing details are omitted.
Main entry points: Training.py Train and scale · Testing.py Inference · MNN_Main.py Storm data split · Data_Processing.py Window and target builder
Components 12
Track preparation
- IBTrACS tracks
ibtracs-tbd.csv - Load and normalize
MNN_Main.py - Storm data split
MNN_Main.py - Window and target builder
Data_Processing.py
Model training
- Train and scale
Training.py - Neural network layer
Add_Layer.py - Model checkpoint
Forecast evaluation
- Inference
Testing.py - Absolute coordinates
MNN_Main.py - Error analysis
MNN_Main.py - Forecast plots
MNN_Main.py
Other components
- Workflow runner
Connections 15
- Workflow runner → Load and normalize: runs workflow
- IBTrACS tracks → Load and normalize: reads tracks
- Load and normalize → Storm data split: normalizes tracks
- Storm data split → Train and scale: training and validation data
- Storm data split → Inference: test data
- Train and scale → Window and target builder: builds windows
- Inference → Window and target builder: builds windows
- Train and scale → Neural network layer: constructs layers
- Inference → Neural network layer: constructs layers
- Train and scale → Model checkpoint: saves best model
- Train and scale → Storm data split: returns scalers
- Inference → Model checkpoint: restores model
- Inference → Absolute coordinates: returns predictions
- Absolute coordinates → Error analysis: computes errors
- Absolute coordinates → Forecast plots: prepares trajectories