neuralODE - Berkeley CS 282 project
report: An empirical study of neural ordinal differential equations
The ODE kernel is based on torchdiffeq, for the details please refer to the README of that Repo.
Image classification tasks
Run
python run_image_classification.py
To compare NeuralODE with ResNet on MNIST / CIFAR10.
Some running examples:
python run_image_classification.py --epochs 200 --dataset cifar10 --batch-size 512 --double "" --network odenet --gpu 1 --lr 0.1 --momentum 0.95 --adjoint 1 --method midpoint
Function fitting tasks
Some running examples:
python run_ode_fitting.py --model spiralNN --viz --adjoint
Time series tasks
Model and code are based on the paper: Deep Multi-Output Forecasting
Running examples:
python multi-output-glucose-forecasting/run.py
Visualization with Visdom
For now visdom can fetch all the csv files following the particular format and plot them.
Go to the Visdom folder then execute the following commands:
visdom -port XXXXX
python visdom_pull_server.py -port XXXXX