1. pandas-cookbook
  2. cookbook
In [2]:
import pandas as pd
pd.set_option('display.mpl_style', 'default')
figsize(15, 3)

We saw earlier that pandas is really good at dealing with dates. It is also amazing with strings! We're going to go back to our weather data from Chapter 5, here.

In [15]:
weather_2012 = pd.read_csv('../data/weather_2012.csv', parse_dates=True, index_col='Date/Time')
weather_2012[:5]
Out[15]:
Temp (C) Dew Point Temp (C) Rel Hum (%) Wind Spd (km/h) Visibility (km) Stn Press (kPa) Weather
Date/Time
2012-01-01 00:00:00 -1.8 -3.9 86 4 8.0 101.24 Fog
2012-01-01 01:00:00 -1.8 -3.7 87 4 8.0 101.24 Fog
2012-01-01 02:00:00 -1.8 -3.4 89 7 4.0 101.26 Freezing Drizzle,Fog
2012-01-01 03:00:00 -1.5 -3.2 88 6 4.0 101.27 Freezing Drizzle,Fog
2012-01-01 04:00:00 -1.5 -3.3 88 7 4.8 101.23 Fog

6.1 String operations

You'll see that the 'Weather' column has a text description of the weather that was going on each hour. We'll assume it's snowing if the text description contains "Snow".

pandas provides vectorized string functions, to make it easy to operate on columns containing text. There are some great examples in the documentation.

In [16]:
weather_description = weather_2012['Weather']
is_snowing = weather_description.str.contains('Snow')

This gives us a binary vector, which is a bit hard to look at, so we'll plot it.

In [17]:
# Not super useful
is_snowing[:5]
Out[17]:
Date/Time
2012-01-01 00:00:00    False
2012-01-01 01:00:00    False
2012-01-01 02:00:00    False
2012-01-01 03:00:00    False
2012-01-01 04:00:00    False
Name: Weather, dtype: bool
In [20]:
# More useful!
is_snowing.plot()
Out[20]:
<matplotlib.axes.AxesSubplot at 0x403c190>

6.2 Use resampling to find the snowiest month

If we wanted the median temperature each month, we could use the resample() method like this:

In [25]:
weather_2012['Temp (C)'].resample('M', how=np.median).plot(kind='bar')
Out[25]:
<matplotlib.axes.AxesSubplot at 0x560cc50>

Unsurprisingly, July and August are the warmest.

So we can think of snowiness as being a bunch of 1s and 0s instead of Trues and Falses:

In [27]:
is_snowing.astype(float)[:10]
Out[27]:
Date/Time
2012-01-01 00:00:00    0
2012-01-01 01:00:00    0
2012-01-01 02:00:00    0
2012-01-01 03:00:00    0
2012-01-01 04:00:00    0
2012-01-01 05:00:00    0
2012-01-01 06:00:00    0
2012-01-01 07:00:00    0
2012-01-01 08:00:00    0
2012-01-01 09:00:00    0
Name: Weather, dtype: float64

and then use resample to find the percentage of time it was snowing each month

In [31]:
is_snowing.astype(float).resample('M', how=np.mean)
Out[31]:
Date/Time
2012-01-31    0.240591
2012-02-29    0.162356
2012-03-31    0.087366
2012-04-30    0.015278
2012-05-31    0.000000
2012-06-30    0.000000
2012-07-31    0.000000
2012-08-31    0.000000
2012-09-30    0.000000
2012-10-31    0.000000
2012-11-30    0.038889
2012-12-31    0.251344
Freq: M, dtype: float64
In [33]:
is_snowing.astype(float).resample('M', how=np.mean).plot(kind='bar')
Out[33]:
<matplotlib.axes.AxesSubplot at 0x5bdedd0>

So now we know! In 2012, December was the snowiest month. Also, this graph suggests something that I feel -- it starts snowing pretty abruptly in November, and then tapers off slowly and takes a long time to stop, with the last snow usually being in April or May.

6.3 Plotting temperature and snowiness stats together

We can also combine these two statistics (temperature, and snowiness) into one dataframe and plot them together:

In [38]:
temperature = weather_2012['Temp (C)'].resample('M', how=np.median)
is_snowing = weather_2012['Weather'].str.contains('Snow')
snowiness = is_snowing.astype(float).resample('M', how=np.mean)

# Name the columns
temperature.name = "Temperature"
snowiness.name = "Snowiness"

We'll use concat again to combine the two statistics into a single dataframe.

In [41]:
stats = pd.concat([temperature, snowiness], axis=1)
stats
Out[41]:
Temperature Snowiness
Date/Time
2012-01-31 -7.05 0.240591
2012-02-29 -4.10 0.162356
2012-03-31 2.60 0.087366
2012-04-30 6.30 0.015278
2012-05-31 16.05 0.000000
2012-06-30 19.60 0.000000
2012-07-31 22.90 0.000000
2012-08-31 22.20 0.000000
2012-09-30 16.10 0.000000
2012-10-31 11.30 0.000000
2012-11-30 1.05 0.038889
2012-12-31 -2.85 0.251344
In [42]:
stats.plot(kind='bar')
Out[42]:
<matplotlib.axes.AxesSubplot at 0x5f59d50>

Uh, that didn't work so well because the scale was wrong. We can do better by plotting them on two separate graphs:

In [43]:
stats.plot(kind='bar', subplots=True, figsize=(15, 10))
Out[43]:
array([<matplotlib.axes.AxesSubplot object at 0x5fbc150>,
       <matplotlib.axes.AxesSubplot object at 0x60ea0d0>], dtype=object)