How yanfei-zhang-95/mnn_cyclonetrack works
MNN_CycloneTrack is a research implementation for predicting cyclone tracks with a Matrix Neural Network. The main workflow loads cyclone training and test sets, trains on sliding windows of track coordinates, predicts test tracks, evaluates mean squared error, and saves predictions and ground truth. Training owns preprocessing, layer construction, and checkpoint access; the main script owns dataset loading and result persistence. A separate testing function evaluates supplied predictions. The README documents the paper but not further product capabilities; the graph reflects the sampled source and listed files.
Main entry points: MNN_Main.py Experiment runner · Training.py Train and predict · Testing.py Prediction testing · Data_processing.py Window preparation
Components 13
Track workflow
- Experiment runner
MNN_Main.py - Cyclone training data
cytrack_train.mat - Cyclone test data
cytrack_test.mat
MNN training
- Train and predict
Training.py - Window preparation
Data_processing.py - MNN layer
Add_layer.py - Model checkpoint
Evaluation and output
- Prediction testing
Testing.py - MSE metric
- Saved predictions
Y_pred_trueMNN.npy - Saved test targets
Y_test.npy - Prior predictions
Y_pred_trueMNN.npy
Other components
- Researcher
Connections 14
- Researcher → Experiment runner: runs
- Cyclone training data → Experiment runner: loads
- Cyclone test data → Experiment runner: loads
- Experiment runner → Train and predict: invokes
- Train and predict → Window preparation: prepares windows
- Train and predict → MNN layer: builds layers
- Train and predict → Model checkpoint: saves and restores
- Train and predict → Experiment runner: returns predictions
- Experiment runner → MSE metric: computes MSE
- Experiment runner → Saved predictions: saves predictions
- Experiment runner → Saved test targets: saves targets
- Prior predictions → Experiment runner: loads
- Prediction testing → Window preparation: prepares targets
- Prediction testing → MSE metric: computes MSE