How to calculate mean using numpy.mean via numpy.ndarray.

sample-max-temperature.py

#!/usr/bin/env python

"""Draw line graph of 118 years August temperature line graph.

Past data of temperature gets from
https://www.data.jma.go.jp/gmd/risk/obsdl/index.php

Sample original data
2018,8,8,30.3,8,2,35.1,8,2,24.8,8,2
2018,8,9,31.2,8,2,37.3,8,2,26.7,8,2
2018,8,10,30.5,8,2,35.4,8,2,27.3,8,2
2018,8,11,29.8,8,2,34.1,8,2,26.1,8,2
2018,8,12,30.4,8,2,34.7,8,2,27.2,8,2

Replace two times
s/(\d{4}),8,(\d{1})/'$1-08-0$3'/
s/(\d{4}),8,(\d{2})/'$1-08-$3'/

Imput data
'2018-08-08',30.3,8,2,35.1,8,2,24.8,8,2
'2018-08-09',31.2,8,2,37.3,8,2,26.7,8,2
'2018-08-10',30.5,8,2,35.4,8,2,27.3,8,2
'2018-08-11',29.8,8,2,34.1,8,2,26.1,8,2
'2018-08-12',30.4,8,2,34.7,8,2,27.2,8,2
"""
from matplotlib import pyplot as plt
from matplotlib import dates as mdates
import numpy as np


dt = np.dtype({
    'names': ['date', 'mean', 'max', 'min'],
    'formats': ['f4', 'f4', 'f4', 'f4']
})

datas = np.loadtxt('data/august-temperature.csv', dtype=dt, skiprows=1,
                   delimiter=',',
                   converters={0: lambda s: mdates.datestr2num(s.decode())},
                   usecols=(0, 1, 4, 7))
# print(datas)

# calculate mean max temperature evey years
BEGIN_YEAR = 1901
END_YEAR = 2018
mdatas = np.ndarray(shape=(1, 4))
mdatas = np.mean(datas[0:31].tolist(), axis=0, keepdims=True)
for year in range(BEGIN_YEAR + 1, END_YEAR):
    # 0 - 30, 31 - 61, 62 - 92,...
    s = (year - BEGIN_YEAR) * 31
    e = s + 31
    mdatas = np.append(mdatas, np.mean(datas[s:e].tolist(), axis=0,
                                       keepdims=True))

# reformat shape(117yeas, 4datas)
mdatas = np.resize(mdatas, (int(len(mdatas)/4), 4))
# Save and read, because I cannot convert labelled ndarray on memory.
np.savetxt('data/mean-temperature.csv', mdatas,
           fmt='%5.1f,%2.2f,%2.2f,%2.2f')
mdatas = np.loadtxt('data/mean-temperature.csv', dtype=dt, skiprows=0,
                    delimiter=',')

datemin = np.datetime64(mdates.num2date(mdatas['date'][0]), 'Y')
datemax = np.datetime64(mdates.num2date(
    mdatas['date'][-1]), 'Y') + np.timedelta64(1, 'Y')
# print(datemax, datemin)

fig, ax = plt.subplots()
ax.set_xlim(datemin, datemax)

# format the cords message box
ax.format_xdata = mdates.DateFormatter('%Y')
ax.format_ydata = lambda x: '$%2.1f' % x  # format the temperature.
ax.grid(True)

ax.plot('date', 'max', data=mdatas, label='max')
ax.plot('date', 'mean', data=mdatas, label='mean')
ax.plot('date', 'min', data=mdatas, label='min')
ax.legend()


# rotates and right aligns the x labels, and moves the bottom of the
# axes up to make room for them
fig.autofmt_xdate()

plt.title('every August mean temperature at Tokyo from 1901 to 2018.')
plt.show()