rtcovidlive / rtlive-global Goto Github PK

Can you please explain better (maybe with some refs), how does exposure variable work?
exposure = pymc3.Deterministic( "exposure", pymc3.math.clip(tests, observed.daily_tests.max() * 0.1, 1e9), dims=["date_with_testcounts"] )
I see here that we are defining a variable "exposure", which is bounded between the maximum number of tests for that day times 0.1 and a huge number.

Later, this variable is used as it follows only for the dates where both case and test count are available (dates present in "mask_exposure"):
exposure = idata.posterior.exposure[:, :, mask_exposure].rename({coord_exposure: "date_with_data"}) exposure_profile = exposure / idata.constant_data.exposure.max()

I do not understand what values can take the "exposure" variable, and moreover how this variable is computed every day. I understood it is similar to the "exposure" present in Poisson regression models, but in doing so aren't we fixing a priori some sort of "rate of positiveness"?
I mean: if we notice that during weekends number of processed tests is way less, and therefore we predict a larger number of them and consequently a larger number of infected people according to this exposure, aren't we overestimating the total number of infected for the very next days? Indeed, the number of processed tests is less during weekends, but these same tests are being processed at the beginning of the next week and will appear in the official data. Aren't we introducing some sort of double counting in doing so?

When importing pymc3 and following instructions written here, it results as the following error.

My OS: Ubuntu 18.04
Python version: 3.8.8 | packaged by conda-forge | (default, Feb 20 2021, 16:22:27)
[GCC 9.3.0]

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-1-cbba610deb25> in <module>
      3 
      4 import datetime
----> 5 import pymc3 as pm
      6 import pandas as pd
      7 import pathlib

~/anaconda3/envs/rtlive/lib/python3.8/site-packages/pymc3/__init__.py in <module>
     62 from .sampling import *
     63 from .smc import *
---> 64 from .stats import *
     65 from .step_methods import *
     66 from .tests import test

~/anaconda3/envs/rtlive/lib/python3.8/site-packages/pymc3/stats/__init__.py in <module>
     45 compare = map_args(az.compare)
     46 ess = map_args(az.ess)
---> 47 geweke = map_args(az.geweke)
     48 hpd = map_args(az.hpd)
     49 loo = map_args(az.loo)

AttributeError: module 'arviz' has no attribute 'geweke'

I think this is a sort of compatibility issue

Impacted states:

• WI
• ...?

Hello!

Good job showing the global data available at https://rtlive.de/global.html

I wonder if you will add the supported regions for each country on that website.

Best regards,
Enzo.

Thanks for the great work you guys do!
I noticed that probably the fedaral states Saarland and Bremen are missing.
If this is not intentional, I think people would be happy to have all states.
Thanks a lot! Dirk

Trying to access the detail page for a nation raise a javascript error:
Uncaught TypeError: Cannot read property 'r_t_threshold_probability' of undefined

Hi,

Thanks for the interesting analysis on rtlive.de!

I realize detailed testcount data is not publicly available for Germany, so it's probably difficult to get a data update.

However, I'm afraid the Prophet forecast is not a great method to get a forecast for current testcount based on 4 week old data. As you can see, rtlive.de indicates a growing number of tests for Germany and the Bundesländer in the past 4 weeks, even though RKI numbers published for those weeks clearly shows the opposite trend: number of tests has been going down in recent weeks. This probably results in underestimating the infection count and R_t.

I'm not sure if there's a good way to fix this unless RKI is willing to give you new data.

Maybe just copy the last week of available data for each Bundesland and scale it according to the change in country-wide test numbers published by RKI for each week? That would probably under-estimate tests in Bavaria and over-estimate tests in other Bundesländer, since Bavaria still has free tests for all, but since there is no public data on a Bundesland level other than https://ars.rki.de/Content/COVID19/Main.aspx (which doesn't appear to be very useful), I don't know a better workaround.

Link for the source in the following notebook:

https://github.com/rtcovidlive/rtlive-global/blob/master/notebooks/Tutorial_model.ipynb

At the section:
##The Bayesian model in PyMC3
The model in PyMC3 follows the above generative process directly as you can see in the source.

points to a link that is not available any more

The data format of the RKI nowcasting XLSX changed.

It is downloaded by https://github.com/rtcovidlive/rtlive-global/blob/master/rtlive/sources/data_de.py#L443

I tried to fix the parsing in https://github.com/rtcovidlive/rtlive-global/blob/master/rtlive/sources/data_de.py#L475 but apparently that didn't work. The RKI no longer shows up in the details plot.