Flip lag parity for python covidcast

[chinandrew]

Realized R and Python are flipped.

in the below examples, note the 3/1 value.

R:

> aggregated = aggregate_signals(list(cases, deaths), dt=list(0,2))
> head(aggregated[order(aggregated$time_value), ])
    geo_value time_value value+0:jhu-csse_confirmed_incidence_prop value+2:jhu-csse_deaths_incidence_prop
213        us 2020-03-01                                 0.0021064                              0.0003009
214        us 2020-03-02                                 0.0069211                              0.0012037
215        us 2020-03-03                                 0.0060184                              0.0003009
216        us 2020-03-04                                 0.0099303                              0.0006018
217        us 2020-03-05                                 0.0231708                              0.0009028
218        us 2020-03-06                                 0.0159487                              0.0012037
> tail(aggregated[order(aggregated$time_value), ])
    geo_value time_value value+0:jhu-csse_confirmed_incidence_prop value+2:jhu-csse_deaths_incidence_prop
130        us 2020-12-27                                  46.82903                              1.0920353
120        us 2020-12-28                                  52.49864                              1.1218263
121        us 2020-12-29                                  60.24400                              1.0375690
122        us 2020-12-30                                  70.28838                              0.6481797
123        us 2020-12-31                                  70.91068                                     NA
131        us 2021-01-01                                  46.22809                                     NA

Python pre-PR:

In [27]: covidcast.aggregate_signals([cases, deaths], dt=[0,2])[['geo_value', 'time_value', 'jhu-csse_confirmed_incidence_prop_0_value', 'jhu-csse_deaths_incidence_prop_1_value']].head()     
Out[27]: 
  geo_value time_value  jhu-csse_confirmed_incidence_prop_0_value  jhu-csse_deaths_incidence_prop_1_value
0        us 2020-03-01                                   0.002106                                     NaN
1        us 2020-03-02                                   0.006921                                     NaN
2        us 2020-03-03                                   0.006018                                0.000000
3        us 2020-03-04                                   0.009930                                0.001505
4        us 2020-03-05                                   0.023171                                0.000301


In [28]: covidcast.aggregate_signals([cases, deaths], dt=[0,2])[['geo_value', 'time_value', 'jhu-csse_confirmed_incidence_prop_0_value', 'jhu-csse_deaths_incidence_prop_1_value']].tail()     
Out[28]: 
    geo_value time_value  jhu-csse_confirmed_incidence_prop_0_value  jhu-csse_deaths_incidence_prop_1_value
304        us 2020-12-30                                  70.288376                                0.601838
305        us 2020-12-31                                  70.910676                                1.092035
306        us 2021-01-01                                  46.228090                                1.121826
307        us 2021-01-02                                        NaN                                1.037569
308        us 2021-01-03                                        NaN                                0.648180

Python with this PR:

In [5]: covidcast.aggregate_signals([cases, deaths], dt=[0,2])[['geo_value', 'time_value', 'jhu-csse_confirmed_incidence_prop_0_value', 'jhu-csse_deat
   ...: hs_incidence_prop_1_value']].head()                                                                                                           
Out[5]: 
  geo_value time_value  jhu-csse_confirmed_incidence_prop_0_value  jhu-csse_deaths_incidence_prop_1_value
0        us 2020-02-28                                        NaN                                0.000000
1        us 2020-02-29                                        NaN                                0.001505
2        us 2020-03-01                                   0.002106                                0.000301
3        us 2020-03-02                                   0.006921                                0.001204
4        us 2020-03-03                                   0.006018                                0.000301

In [6]: covidcast.aggregate_signals([cases, deaths], dt=[0,2])[['geo_value', 'time_value', 'jhu-csse_confirmed_incidence_prop_0_value', 'jhu-csse_deat
   ...: hs_incidence_prop_1_value']].tail()                                                                                                           
Out[6]: 
    geo_value time_value  jhu-csse_confirmed_incidence_prop_0_value  jhu-csse_deaths_incidence_prop_1_value
304        us 2020-12-28                                  52.498642                                1.121826
305        us 2020-12-29                                  60.243998                                1.037569
306        us 2020-12-30                                  70.288376                                0.648180
307        us 2020-12-31                                  70.910676                                     NaN
308        us 2021-01-01                                  46.228090                                     NaN

[capnrefsmmat]

Looks good, just one suggested improvement to make your improved docs even more improved

[dshemetov]

While this looks like the direction @capnrefsmmat @chinandrew were happy to move a couple years ago, this would break a lot of downstream code today, as it's a backwards-incompatible change. Motion to close this PR?

[capnrefsmmat]

Agreed -- too late to change it now.