Johns Hopkins COVID-19 Dataset in Pandas

COVID-19 is ravaging the globe. Let’s look at the excellent Johns Hopkins dataset using Pandas. This will serve both as a guideline for getting the data and exploring on your own, as well as an example of Pandas multi-indexing in an easy to understand situation. I am currently involved in science-responds.

My favorite links: worldometerarcGISProjections

A few more: COVID-19 dashnCoV201991-DIVOC

# Python 3.6+ required
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from urllib.error import HTTPError
from typing import Optional
plt.style.use('ggplot')

Anyway, now that we’ve made some basic imports, let’s write a function that can read in a datafile from GitHub:

def valid_name(name: str) -> bool:
    "Return True if this is a valid name in the dataset. Faster parsing this way."
    
    return name.replace("/", "_").replace(" ", "_") in {
        "Country_Region", "Province_State", "Admin2", "Confirmed", "Deaths", "Recovered"
    }
def get_day(day: pd.Timestamp):

    # Read in a datafile from GitHub
    try:
        table = pd.read_csv(
            "https://raw.githubusercontent.com/CSSEGISandData/COVID-19/"
            "master/csse_covid_19_data/csse_covid_19_daily_reports/"
            f"{day:%m-%d-%Y}.csv",
            usecols=valid_name,
        )
    except HTTPError:
        return pd.DataFrame()
    
    # Cleanup - sadly, the format has changed a bit over time - we can normalize that here
    table.columns = [
        f.replace("/", "_")
        .replace(" ", "_")
        for f in table.columns
    ]
    
    # This column is new in recent datasets
    if "Admin2" not in table.columns:
        table["Admin2"] = None
    
    # If the last update time was useful, we would make this day only, rather than day + time
    #   table["Last_Update"] = pd.to_datetime(table["Last_Update"]).dt.normalize()
    #
    # However, last update is odd, let's just make this the current day
    table["Last_Update"] = day
    
    # Make sure indexes are not NaN, which causes later bits to not work. 0 isn't
    # perfect, but good enough.
    # Return as a multindex
    return table.fillna(0).set_index(
        ["Last_Update", "Country_Region", "Province_State", "Admin2"], drop=True
    )

Now let’s loop over all days and build a multi-index DataFrame with the whole dataset. We’ll be doing quite a bit of cleanup here as well. If you do this outside of a function, you should never modify an object in multiple cells; ideally you create an object like df, and make any modifications and replacements in the same cell. That way, running any cell again or running a cell multiple times will not cause unusual errors and problems to show up.

def get_all_days(end_day: Optional[pd.Timestamp] = None) -> pd.DataFrame:

    # Assume current day - 1 is the latest dataset if no end given
    if end_day is None:
        end_day = pd.Timestamp.now().normalize()

    # Make a list of all dates
    date_range = pd.date_range("2020-01-22", end_day)
    
    # Create a generator that returns each day's dataframe
    day_gen = (get_day(day) for day in date_range)
    
    # Make a big dataframe, NaN is 0
    df = pd.concat(day_gen).fillna(0).astype(int)
    
    # Remove a few duplicate keys
    df = df.groupby(level=df.index.names).sum()
    
    # Sometimes active is not filled in; we can compute easily
    df["Active"] = np.clip(
        df["Confirmed"] - df["Deaths"] - df["Recovered"], 0, None
    )
    
    # Change in confirmed cases (placed in a pleasing location in the table)
    df.insert(
        1,
        "ΔConfirmed",
        df.groupby(level=("Country_Region", "Province_State", "Admin2"))["Confirmed"]
        .diff()
        .fillna(0)
        .astype(int),
    )
    
    # Change in deaths
    df.insert(
        3,
        "ΔDeaths",
        df.groupby(level=("Country_Region", "Province_State", "Admin2"))["Deaths"]
        .diff()
        .fillna(0)
        .astype(int),
    )
    
    return df

If this were a larger/real project, it would be time to bundle up the functions above and put them into a .py file - notebooks are for experimentation, teaching, and high level manipulation. Functions and classes should normally move to normal Python files when ready.

Let’s look at a few lines of this DataFrame to see what we have:

df = get_all_days()
df
Confirmed ΔConfirmed Deaths ΔDeaths Recovered Active
Last_Update Country_Region Province_State Admin2
2020-01-22 China Unknown 0 0 0 0 0 0 0
Hong Kong Hong Kong 0 0 0 0 0 0 0
Japan 0 0 2 0 0 0 0 2
Kiribati 0 0 0 0 0 0 0 0
Macau Macau 0 1 0 0 0 0 1
... ... ... ... ... ... ... ... ... ...
2021-06-23 Vietnam 0 0 13989 207 70 1 5684 8235
West Bank and Gaza 0 0 313015 0 3555 0 306532 2928
Yemen 0 0 6898 6 1355 0 3990 1553
Zambia 0 0 137026 3367 1794 50 113109 22123
Zimbabwe 0 0 43480 766 1692 1 37477 4311

1767078 rows × 6 columns

The benefit of doing this all at once, in one DataFrame, should quickly become apparent. We can now use simple selection and grouping to “ask” almost anything about our dataset.

As an example, let’s look at just the US portion of the dataset. We’ll use the pandas selection .xs:

us = df.xs("US", level="Country_Region")
us
Confirmed ΔConfirmed Deaths ΔDeaths Recovered Active
Last_Update Province_State Admin2
2020-01-22 Washington 0 1 0 0 0 0 1
2020-01-23 Washington 0 1 0 0 0 0 1
2020-01-24 Chicago 0 1 0 0 0 0 1
Washington 0 1 0 0 0 0 1
2020-01-25 Illinois 0 1 0 0 0 0 1
... ... ... ... ... ... ... ... ...
2021-06-23 Wyoming Teton 3798 0 11 0 0 3787
Uinta 2304 -4 13 0 0 2291
Unassigned 0 0 0 0 0 0
Washakie 927 0 26 0 0 901
Weston 656 0 6 0 0 650

1468158 rows × 6 columns

Notice we have counties (early datasets just have one “county” called "0"). If we were only interested in states, we can group by the remaining levels and sum out the "Admin2" (county and similar) dimension:

by_state = us.groupby(level=("Last_Update", "Province_State")).sum()
by_state
Confirmed ΔConfirmed Deaths ΔDeaths Recovered Active
Last_Update Province_State
2020-01-22 Washington 1 0 0 0 0 1
2020-01-23 Washington 1 0 0 0 0 1
2020-01-24 Chicago 1 0 0 0 0 1
Washington 1 0 0 0 0 1
2020-01-25 Illinois 1 0 0 0 0 1
... ... ... ... ... ... ... ...
2021-06-23 Virginia 679137 228 11368 1 0 667769
Washington 448945 803 5889 46 0 443056
West Virginia 163689 62 2872 2 0 160817
Wisconsin 677048 117 8077 10 0 668971
Wyoming 61776 72 740 6 0 61036

28190 rows × 6 columns

Using the same selector as before, we can pick out North Carolina:

by_state.xs("North Carolina", level="Province_State")
Confirmed ΔConfirmed Deaths ΔDeaths Recovered Active
Last_Update
2020-03-10 7 0 0 0 0 7
2020-03-11 7 0 0 0 0 7
2020-03-12 15 8 0 0 0 15
2020-03-13 17 2 0 0 0 17
2020-03-14 24 7 0 0 0 24
... ... ... ... ... ... ...
2021-06-19 1010113 0 13340 0 0 996773
2021-06-20 1010113 0 13340 0 0 996773
2021-06-21 1010889 776 13368 28 0 997521
2021-06-22 1011100 211 13382 14 0 997718
2021-06-23 1011561 461 13393 11 0 998168

471 rows × 6 columns

We can look at all of US, as well:

all_states = by_state.groupby(level="Last_Update").sum()
all_states
Confirmed ΔConfirmed Deaths ΔDeaths Recovered Active
Last_Update
2020-01-22 1 0 0 0 0 1
2020-01-23 1 0 0 0 0 1
2020-01-24 2 0 0 0 0 2
2020-01-25 2 0 0 0 0 2
2020-01-26 5 0 0 0 0 5
... ... ... ... ... ... ...
2021-06-19 33537995 8520 601741 170 0 32939414
2021-06-20 33541887 3892 601824 83 0 32943222
2021-06-21 33554275 12388 602092 268 0 32955291
2021-06-22 33565215 10940 602462 370 0 32965867
2021-06-23 33577651 12436 602837 375 0 32977928

519 rows × 6 columns

US total cases

Let’s try a simple plot first; this is the one you see quite often.

plt.figure(figsize=(10,5))
all_states.Confirmed.plot(logy=True, style='o');

png

Italy, new cases per day

As another example, let’s view the new cases per day for Italy. We will add a rolling mean, just to help guide the eye through the fluctuations - it is not a fit or anything fancy.

interesting = df.xs("Italy", level="Country_Region").groupby(level="Last_Update").sum()

plt.figure(figsize=(10,5))
interesting.ΔConfirmed.rolling(5, center=True).mean().plot(style='-', label='Rolling mean')
interesting.ΔConfirmed.plot(style='o', label="Data")
plt.ylabel("New cases per day")
plt.legend();

png

Italy, transmission rate

It’s more interesting to instead look at the transmission rate per day, which is new cases / active cases. The colors in the plot start changing when Italy implemented a lockdown on the 11th, and change over 14 days, which is roughly 1x the time to first symptoms. The lockdown make take longer than that to take full effect. There were several partial steps taken before the full lockdown on the 4th and 9th. Notice the transmission is slowing noticeably!

interesting = df.xs("Italy", level="Country_Region").groupby(level="Last_Update").sum()
growth = interesting.ΔConfirmed / interesting.Active
growth = growth['2020-02-24':]

# Color based on lockdown (which happened in 3 stages, 4th, 9th, and 11th)
lockdown = growth.index - pd.Timestamp('2020-03-11')
lockdown = np.clip(lockdown.days, 0, 14) / 14

fix, ax = plt.subplots(figsize=(10,5))
ax.scatter(growth.index, growth, cmap='cool', c=lockdown)

ax.set_ylabel("new cases / active cases")
Text(0, 0.5, 'new cases / active cases')

png

US, transmission rate

Same plot for the US. The colors in the plot start changing when the US started the 15 plan to slow the spread, and change over 14 days, which is roughly 1x the time to first symptoms. Each state has implemented different guidelines, so the effect will be spread out even futher. Again, we are see the effect of the lockdown!

interesting = df.xs("US", level="Country_Region").groupby(level="Last_Update").sum()
growth = interesting.ΔConfirmed / interesting.Active
growth = growth['2020-03-01':]

# Not really a full lockdown, just a distancing guideline + local lockdowns later
lockdown = growth.index - pd.Timestamp('2020-03-15')
lockdown = np.clip(lockdown.days, 0, 14) / 14

fix, ax = plt.subplots(figsize=(10,5))
ax.scatter(growth.index, growth, cmap='cool', c=lockdown)

ax.set_ylabel("new cases / active cases")
Text(0, 0.5, 'new cases / active cases')

png


comments powered by Disqus