search

lisphilar / covid19-sir

CovsirPhy: Python library for COVID-19 analysis with phase-dependent SIR-derived ODE models.
https://lisphilar.github.io/covid19-sir/
Apache License 2.0
110 stars 44 forks source link

[Discuss] Questions about the SIRF hyperparameter optimization example #394

Closed Inglezos closed 3 years ago

Inglezos commented 3 years ago

Summary of question

  1. For SIRF Hyperparameter optimization example, why are the estimated parameters so much different than the expected? Is this bad or this is simply just another set of solutions for the parameters, which lead to the same case results? I get for theta = 0.0195 (expected is 0.002) and for kappa = 0.0032 (expected is 0.005).

  2. What does the trajectory plot exactly mean? I see that there is not a single solution set. For this plot I get: sirf_example_parameters

but you have something totally different in the kaggle notebook results for SIRF example: image

Inglezos commented 3 years ago

As I noticed just now in the latest kaggle notebook, the plot there is the same now with ours, the seemingly wrong one.

lisphilar commented 3 years ago

We did not change the codes related to parameter estimation and scripts of the section of the Kaggle Notebook.... Could updating dependencies impact on them...?

lisphilar commented 3 years ago

This is not a problem in itself, cut-off of long estimation could be implemented.

Inglezos commented 3 years ago

So what caused the different behavior? The long runtime?

lisphilar commented 3 years ago

Simply Optuna is strugging to find out the best parameter values. The graph is normal. We need to give Optuna hints to find the best params with new ideas and end interation when RMSLE score does not change for the last iterations.