# Prediction of Danish Covid19 cases

**URL:** <https://discourse.pymc.io/t/prediction-of-danish-covid19-cases/4904>\
**Category:** Questions\
**Created:** [April 20, 2020, 11:28am UTC](https://discourse.pymc.io/t/prediction-of-danish-covid19-cases/4904 "2020-04-20T11:28:38Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![sojohan](https://avatars.discourse-cdn.com/v4/letter/s/ecb155/32.png) [@sojohan](https://discourse.pymc.io/u/sojohan)\
**Post date:** [April 20, 2020, 11:28am UTC](https://discourse.pymc.io/t/prediction-of-danish-covid19-cases/4904/1 "2020-04-20T11:28:38Z")

</div>

Hi Experts

I have made a simple model for predicting the number of Danish Covid19 Hospitalized patients with Gaussian Processes.

My notebook is here:  
[[https://github.com/sojohan/Prediction-of-Danish-COVID-19-Cases/blob/master/GaussianProcess\_Danish\_Corona\_figures.ipynb](https://github.com/sojohan/Prediction-of-Danish-COVID-19-Cases/blob/master/GaussianProcess_Danish_Corona_figures.ipynb)]

Would be very happy for feedback?

Sojohan

---

<div class="post-metadata">

**Author:** ![fonnesbeck](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/fonnesbeck/32/13_2.png) [@fonnesbeck](https://discourse.pymc.io/u/fonnesbeck)\
**Post date:** [April 27, 2020, 7:03pm UTC](https://discourse.pymc.io/t/prediction-of-danish-covid19-cases/4904/2 "2020-04-27T19:03:58Z")

</div>

Hey Sojohan,

GPs are generally my go-to approach for non-linear time series, but the reservation I have in this case is that it is not a mechanistic model, so I would be hesitant to use it in place of S(E)IR compartmental models. A GP model might be useful for very short-term predictions (depending on the length scale), but since it does not account for changes in behavior (including specific interventions), I think it has limited use. That said, it might be interesting to incorporate a GP into a mechanistic model as a way of estimating some of the latent parameters, and their dynamics.

One technical observation: once you have fit the model using MCMC (via `pm.sample`) there is no need to fit it again using `find_MAP`. The latter is actually an inferior way of fitting the model, since it is not fully Bayesian–it just does brute-force mode hunting.

Chris

---

<div class="post-metadata">

**Author:** ![sojohan](https://avatars.discourse-cdn.com/v4/letter/s/ecb155/32.png) [@sojohan](https://discourse.pymc.io/u/sojohan)\
**Post date:** [April 28, 2020, 8:31am UTC](https://discourse.pymc.io/t/prediction-of-danish-covid19-cases/4904/3 "2020-04-28T08:31:36Z")

</div>

Hi Chris

Thanks for the feedback. Yes I agree that the model is good at short-term predictions. Would try to update priors and see the results.

Yes I know that find\_map was not necessary…

sojohan