covid-growth.py

# Based on this repository of coronavirus data:
#   git clone [email protected]:nytimes/covid-19-data.git

from csv import DictReader
from collections import defaultdict
from pprint import pprint

# fields = ['date', 'state', 'cases', 'deaths']
# Selecting on 'cases' gives results that appear influenced by testing.
# Using 'deaths' appears less influenced by noise and human factors, in my opinion.
column_name = 'deaths'
# column_name = 'cases'
with open('us-states.csv') as handle:
    dd = DictReader(handle)
    d = defaultdict(dict)
    for row in dd:
        d[row['state']].update({row['date']: int(row[column_name])})

# Cumulative cases sorted by date for each state
cases = {
    state: [v for k, v in sorted(cases.items(), key=lambda x: x[0])]
    for state, cases in d.items()
}

# 5-day moving average of day-over-day rates by state
n = 5
rate_ma = {
    state: [
        (state_cases[i + n] / float(state_cases[i])) ** (1. / n)
        for i in range(len(state_cases) - n)
        if state_cases[i] >= 10
    ]
    for state, state_cases in cases.items()
}

pprint(
    [
        (k, v and v[-1])
        for k, v in sorted(
            ((k, v) for k, v in rate_ma.items() if v),
            key=lambda x: x[1][-1]
        )
    ]
)

5-day moving geometric average of change in rate of death (as of 2020-03-27):

[('Washington', 1.0971112499422204),
 ('Florida', 1.213950735423196),
 ('Georgia', 1.2286596790831472),
 ('California', 1.2397415710064692),
 ('New Jersey', 1.3831618672225916),
 ('Louisiana', 1.3899258176361557),
 ('New York', 1.4011310323534027)]