use hospitalizations data in place of deaths

This file should not have any jax dependencies.

The goal is to make it so that someone without jax installed can still do data loading and cleaning.

we want to free up the H name for the hospitalizations pathway

to start, only support US states and national, getting data from the covidcast API.

For states:

• rename jhu to something more general
• add argument to jhu.load_us to specify which data source to load from (jhu or covidcast)
• add a utility function similar to load_us_time_series that gets cases, hospitalizations, or deaths data from covidcast for all states. Function arguments are target_variable ("cases", "hospitalizations", or "deaths"), and an optional as_of parameter. It should return a pandas data frame with the date as an index and one column for each state-level location. The column names should match the location names returned by jhu.get_state_info(). The values in the columns will be the case/hospitalization/death counts (just one for a single call to the function) in each state on the given date.
• At

  mechbayes/mechbayes/jhu.py

  Lines 145 to 146 in c51d090

  	confirmed = load_us_time_series("time_series_covid19_confirmed_US.csv")
  	deaths = load_us_time_series("time_series_covid19_deaths_US.csv")

  , condition on the data source to either do the existing calls to load_us_time_series or call the new function above. For now, when calling the utility function for covidcast, just omit the as_of parameter to get the latest available data, matching current behavior for cases and deaths. If data source is covidcast, add hosps to the concat call a couple lines down.

For US national:

• add an entry to the dictionary at

  mechbayes/mechbayes/jhu.py

  Line 26 in c51d090

  d = {key: load_and_massage(url) for key, url in sources.items()}

  for hosp. For now, it should have observed value for the US only, and nan's for all other locations

For counties:

• we don't have observations for counties, add a column with all nan's