Fix issue 255 #273

quinn-dougherty · 2020-03-26T19:14:33Z

Closes #255

found a clean way to do it in streamlit's UI that doesn't build up ugly technical debt.

TODO: need to rearrange the order in which computations happen in order to get it to actually work. in the notebook, inferred doubling time is a function of assumed doubling time, so i just need to make that work with autopopulating the form with a default (and not write myself into literal circles)

src/penn_chime/presentation.py

quinn-dougherty · 2020-03-26T19:54:51Z

inferred doubling time is a function of assumed doubling time

I was mistaken about this, sorry, correcting approach.

quinn-dougherty · 2020-03-26T21:17:58Z

Re Phil's comment. (streamlit's st.date_input)

cjbayesian · 2020-03-26T21:47:47Z

@quinn-dougherty That looks perfect. Just fix the typo -> "I know the date of the first [...]" (remove 'number of')

src/penn_chime/models.py

PhilMiller · 2020-03-26T22:04:57Z

src/penn_chime/models.py

+        if p.n_days_since_first_hospitalized is not None:
+            # do not shorten this to `if foo:` form. if you do and n_days_since_first_hospitalized is 0,
+            # program will crash.
+            p.doubling_time = self.argmin_dt(np.linspace(1, 15, 29), p)


This is a really horrifying thought, but is there a principled reason the doubling time can never be less than 1 day? Super-spreader events would seem to do exactly that.

that's a hardcoded constant from Corey's notebook that we can factor into the settings.cfg file if we wish. since we're pushing to prod weekly, we only hope that things don't get that drastic too quickly ;)

More broadly, if the parameter sweep from argmin_dt returns a value at either extreme of its range, it means either

The best fit value is approximately the given limit

The fit failed because the true result is outside the range

We need to not let the latter case go through without warning or examination. It may just as well be an indicator of mis-keyed inputs, as of the math honestly saying anything about the circumstances.

There's a physical limit on doubling time down around 0.1, since that implies 10 daily doublings, or about 1000x, which exhausts itself pretty damn quickly. I think there's some threshold higher than that where we should warn the user that the computed value is implausibly low

i think you're right. i'll add in some sort of warning system.

I believe that a different merge allowed doubling time to be a float/fractional.

src/penn_chime/models.py

PhilMiller · 2020-03-26T23:13:15Z

A UI note for as this progresses: The number of days for which the user asks for a projection needs to be in addition to days in the back-cast - i.e. the admissions and census charts should run from -1*n_days_since_first_hospitalized to n_days

quinn-dougherty · 2020-03-26T23:47:31Z

@cjbayesian
your notebook has S, I, R = population_size-initial_i, initial_i, 0, your susceptible population is total_population - initial_infected.

This is not parity with prod. The codebase considers the entire total_population susceptible.

how should we proceed? initial_i (infected as we call it in prod) is 3 orders of magnitude smaller than total population, so it shouldn't be effecting our results right now.

PhilMiller · 2020-03-26T23:54:10Z

That would be #211, which I realized I need to go back and revise the fix in #212 for the big refactoring.

cjbayesian · 2020-03-27T11:54:54Z

Yep, @PhilMiller is right that I had fixed the susceptible calculation in the notebook per issue #211 .

quinn-dougherty · 2020-03-27T12:05:32Z

That would be #211, which I realized I need to go back and revise the fix in #212 for the big refactoring.
Yep, @PhilMiller is right that I had fixed the susceptible calculation in the notebook per issue #211.

#276 merged

cjbayesian · 2020-03-27T12:09:55Z

@PhilMiller and @quinn-dougherty : I have a coherent way to handle either case of user input (doubling_time or n_days_since_first_hospitalized) which addresses also #187 .

I just pushed an update to my notebook showing how. It's basically the same fitting procedure, just holding doubling_time constant and fitting n_days_since_first_hospitalized instead.

into quinn_issue_255

quinn-dougherty · 2020-03-27T12:45:00Z

Per Phil's comments, someone remind us to have a talk about what this np.linspace(1,15,29) should be set to before push to prod monday.

quinn-dougherty · 2020-03-27T13:04:59Z

not using vectorized argmin because i'm keeping track of state.
changed obs_pred_diff to loss_dt

pushing some commits without the 187 piggyback shortly

quinn-dougherty · 2020-03-27T14:53:19Z

I have a coherent way to handle either case of user input (doubling_time given n_days_since_first_hospitalized) which addresses also #187 .
I just pushed an update to my notebook showing how. It's basically the same fitting procedure, just holding doubling_time constant and fitting n_days_since_first_hospitalized instead.

sorry, this is gonna have to be another PR. racing to get the basics of 255 into Prod monday.

quinn-dougherty · 2020-03-27T15:40:55Z

there's a math error -- Corey's notebook infers a doubling time of 4 with similar initial conditions, my implementation infers a doubling time of 15 :(

I'll also recall Phil's warning, it's reporting the 15th day (the max of the range) as the doubling time on input 03-09-2020 (with checkbox "i know date of first hospitalized")

what Phil wrote:

More broadly, if the parameter sweep from argmin_dt returns a value at either extreme of its range, it means either

The best fit value is approximately the given limit
The fit failed because the true result is outside the range
We need to not let the latter case go through without warning or examination. It may just as well be an indicator of mis-keyed inputs, as of the math honestly saying anything about the circumstances.

quinn-dougherty · 2020-03-27T15:43:39Z

TODO:

implement warning if inferred doubling time is at the top or the bottom of its search space.

jlubken · 2020-03-27T15:48:25Z

src/penn_chime/models.py

+                recovered,
+                beta,
+                gamma,
+                p.n_days


Shouldn't this be p.n_days + p.n_days_since_first_hospitalization?

all i'll say is that wouldn't be parity with the notebook. @cjbayesian

OK, so we'll fix this later then. We really want the raw_df to hold all of the data instead of having to bodge n_days_since... into every function.

I'll get a copy of the notebook and test for equivalent output later.

However, this input can no longer be described as the number of days to project as stated in the ui. It is projecting less than n_days, it is projecting from the first infection. The ui / users here will be confusing / confused.

quinn-dougherty · 2020-03-27T15:52:25Z

hi -- are we ok if i piggyback updating settings.py in here? i'm trying to achieve parity with the notebook. i'm assuming it'll happen before monday anyway.

… into my_more_issue_255

Put social distancing effect back in. Starts at current day.

These are irrelevent to the calculations and only cause user confusion.

Remove confirmed cases & detection prob

Add time as a parameter

first push of issue 255

7e39b63

mdbecker requested review from PhilMiller and cjbayesian March 26, 2020 19:43

mdbecker assigned quinn-dougherty Mar 26, 2020

PhilMiller reviewed Mar 26, 2020

View reviewed changes

src/penn_chime/presentation.py Outdated Show resolved Hide resolved

it now asks user date of first hospitalization

11c6e2a

PhilMiller reviewed Mar 26, 2020

View reviewed changes

src/penn_chime/models.py Outdated Show resolved Hide resolved

PhilMiller reviewed Mar 26, 2020

View reviewed changes

src/penn_chime/models.py Outdated Show resolved Hide resolved

Merge branch 'develop' into quinn_issue_255

0a978ff

quinn-dougherty added 2 commits March 27, 2020 08:09

incremental progress

a3d1a9a

Merge branch 'quinn_issue_255' of https://github.com/CodeForPhilly/chime

1404983

into quinn_issue_255

code is in, have to reset x axis

e9b4fbf

day zero of graph is now the user inputted date_first_hospitalized

124a3fd

jlubken reviewed Mar 27, 2020

View reviewed changes

quinn-dougherty and others added 22 commits March 29, 2020 09:27

Merge branch 'more_issue_255' of https://github.com/CodeForPhilly/chime…

45c55dc

… into my_more_issue_255

Remove duplicate computation

25bdbb0

Replace n_days_since with i_day

bd1d060

Put social distancing effect back in. Starts at current day.

ce6fa6c

More logging of givens and estimates

392e907

Log len df and expected range

57f3b99

Resolve changes from upstream

a562191

Fix test reflecting social distancing

2bdf133

Add social distancing

e160cbc

Put social distancing effect back in. Starts at current day.

Move n_days back to 60 so social distancing can be seen in the plots

ef8e99b

Remove confirmed cases & detection prob

ba8d23a

These are irrelevent to the calculations and only cause user confusion.

Update change_date notice

d2d06f9

Copy params for tests

38c2901

Add copy for "constant" PARAM to MODEL and HALVING_MODEL

7755807

Fix tests for r_t

da33480

Remove detection prob

5c3d3b6

Remove confirmed cases & detection prob

Patch all

5dab315

Add settings

3e7612c

Try to fix cypress

fa2f740

Off by one for cypress

6e4b2d2

Add lots of time

c8cad06

Add time as a parameter

Resolve issue 255

0eed3c6

jlubken approved these changes Mar 29, 2020

View reviewed changes

jlubken changed the title ~~WIP issue 255~~ Fix issue 255 Mar 29, 2020

jlubken merged commit 76dd9b9 into develop Mar 29, 2020

jlubken deleted the quinn_issue_255 branch March 29, 2020 17:07

quinn-dougherty mentioned this pull request Mar 29, 2020

Census should not start from 0 #187

Closed

PhilMiller mentioned this pull request Mar 30, 2020

["model"] Check impact of latent period delay from infectious to hospitalized on model dynamics and forecasts #340

Open

quinn-dougherty mentioned this pull request Mar 30, 2020

["dash"] catchup to pull 273 (v1.1) #374

Closed

PhilMiller mentioned this pull request Mar 30, 2020

["model"] Fitting to current census with a long projection can match to the down-slope #378

Closed

Fix issue 255 #273

Fix issue 255 #273

Uh oh!

Conversation

quinn-dougherty commented Mar 26, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

quinn-dougherty commented Mar 26, 2020

Uh oh!

quinn-dougherty commented Mar 26, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cjbayesian commented Mar 26, 2020

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

PhilMiller commented Mar 26, 2020

Uh oh!

quinn-dougherty commented Mar 26, 2020

Uh oh!

PhilMiller commented Mar 26, 2020

Uh oh!

cjbayesian commented Mar 27, 2020

Uh oh!

quinn-dougherty commented Mar 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cjbayesian commented Mar 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

quinn-dougherty commented Mar 27, 2020

Uh oh!

quinn-dougherty commented Mar 27, 2020

Uh oh!

quinn-dougherty commented Mar 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

quinn-dougherty commented Mar 27, 2020

Uh oh!

quinn-dougherty commented Mar 27, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

quinn-dougherty commented Mar 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

quinn-dougherty commented Mar 26, 2020 •

edited

Loading

quinn-dougherty commented Mar 26, 2020 •

edited

Loading

quinn-dougherty commented Mar 27, 2020 •

edited

Loading

cjbayesian commented Mar 27, 2020 •

edited

Loading

quinn-dougherty commented Mar 27, 2020 •

edited

Loading

quinn-dougherty commented Mar 27, 2020 •

edited

Loading