Machine learning methods applied to the SARCOS dataset, treated as a multivariate regression problem.
Note: As discovered by @rajshah4, there is a large amount of leakage between the training and test sets. This code and results are currently left as-is for reference, but should not be considered representative of results with a proper training/test split.
Method | MSE | # Params | # Avg. Path Params |
---|---|---|---|
Linear regression | 10.69263 | 154 | N/A |
Decision tree | 3.70763 | 319,591 | 24.6 |
Neural network (1 hidden layer) | 2.83472 | 7,431 | N/A |
Neural network (5 hidden layers) | 2.65670 | 270,599 | N/A |
Random forest | 2.39401 | 141,540,436 | 16,771.0 |
Neural network (3 hidden layers) | 2.12862 | 139,015 | N/A |
Gradient boosted trees | 1.44412 | 988,256 | 6,807.7 |