Resample (and some other cool stuff) demo!

Looking at the Bellingham weather data, I'm going to look at the HourlyRelativeHumidity (mostly because I don't have to clean it) and resample it to examine it on a per-month level. Start by loading the data:

import pandas as pd

def data_url(city_name):
    base_url = "https://facultyweb.cs.wwu.edu/~wehrwes/courses/csci141_21s/fp/data/"
    return base_url + city_name + ".csv"

bham = pd.read_csv(data_url("WA_Bellingham"), parse_dates=["DATE"], low_memory=False)

Here's the magic of resample in action. What's happening here?

• The first argument "MS" says to resample Monthly, labeling each group with the Start of the month.
• on="DATE" is the column containing the timestamps we're resampling on This resample call resamples the entire dataframe, all columns included. Since we're just intersted in RH, we'll get only that column from the result.

# this gives us the maximum hourly relatiev humidity
hourly_humidity_permonth = bham.resample("MS", on="DATE")["HourlyRelativeHumidity"]
hourly_humidity_permonth

<pandas.core.groupby.generic.SeriesGroupBy object at 0x16c45e190>

So that's interesting! The result of a resample is a familiar GroupBy object - the same thing we get when we call groupby. That means we can't directly poke at the results (it's like a pile of multiple dataframes), but we can call pandas functions on it to aggregate each one. What's the maximum RH per month? Just call max() on the groupby result.

hourly_humidity_permonth.max()

DATE
2020-01-01    100.0
2020-02-01    100.0
2020-03-01    100.0
2020-04-01     94.0
2020-05-01     94.0
2020-06-01     94.0
2020-07-01     94.0
2020-08-01     97.0
2020-09-01    100.0
2020-10-01    100.0
2020-11-01     97.0
2020-12-01    100.0
Freq: MS, Name: HourlyRelativeHumidity, dtype: float64

Average RH per month?

hourly_humidity_permonth.mean()

DATE
2020-01-01    76.605477
2020-02-01    79.876310
2020-03-01    70.918979
2020-04-01    65.603015
2020-05-01    69.106618
2020-06-01    73.459698
2020-07-01    72.490196
2020-08-01    72.806373
2020-09-01    81.267051
2020-10-01    82.047081
2020-11-01    79.328092
2020-12-01    89.361538
Freq: MS, Name: HourlyRelativeHumidity, dtype: float64

Want all the summary stats? describe is your tool of choice:

hourly_humidity_permonth.describe()

	count	mean	std	min	25%	50%	75%	max
DATE
2020-01-01	986.0	76.605477	10.910598	27.0	70.00	79.0	85.0	100.0
2020-02-01	954.0	79.876310	11.092164	39.0	73.00	82.0	89.0	100.0
2020-03-01	901.0	70.918979	17.570527	22.0	61.00	75.0	85.0	100.0
2020-04-01	796.0	65.603015	16.913935	16.0	55.75	68.0	79.0	94.0
2020-05-01	816.0	69.106618	15.069562	28.0	58.00	72.0	80.0	94.0
2020-06-01	794.0	73.459698	11.880193	43.0	65.00	75.0	83.0	94.0
2020-07-01	816.0	72.490196	13.332921	34.0	63.00	75.0	84.0	94.0
2020-08-01	816.0	72.806373	13.934968	27.0	61.00	75.0	84.0	97.0
2020-09-01	1041.0	81.267051	14.993581	18.0	75.00	87.0	93.0	100.0
2020-10-01	1062.0	82.047081	12.306165	37.0	75.25	86.0	92.0	100.0
2020-11-01	954.0	79.328092	12.254657	25.0	74.00	82.5	89.0	97.0
2020-12-01	1040.0	89.361538	9.785872	44.0	85.00	90.0	97.0	100.0

Of course, I can't resist any excuse I can find to make a histogram. Remember that we can loop over the results of a groupy (and therefore the results of a resample, too). So let's histogram each month. I looked up a few Seaborn tricks to make this nicer:

• I'm making sure the bins are consistent across all months, and including all 10 bins of size 10 even if there's no data there.
• I'm setting the title of each plot to the name of the month, which if you can get from a DateTime object if you ask nicely (using the month_name() function).

import seaborn as sns
import matplotlib.pyplot as plt

for month, data in hourly_humidity_permonth:
    sns.histplot(data, bins=range(0, 110, 10)).set(title=month.month_name())
    plt.show()

What the heck, now I want to know what Arizona looks like. In case the urge to look at any other cities strikes, I'm going to write a function that can be called on any city's data to generate these plots.

def humidity_plots(df):
    resampled = df.resample("MS", on="DATE")["HourlyRelativeHumidity"]
    
    for month, data in resampled:
        sns.histplot(data, bins=range(0, 110, 10)).set(title=month.month_name())
        plt.show()

phoenix = pd.read_csv(data_url("AZ_Phoenix"), low_memory=False, parse_dates=["DATE"])

humidity_plots(phoenix)

Revealed: Phoenix is a lot drier than Bellingham!