Total of pip packages downloaded, separated by Python versions

README.md

Total of pip packages downloaded separated by Python versions

From June 26, 2016 (python 3.5.2 release) to Aug. 31, 2016.

Python versions from 2.6 to 3.5

Without 2.7

###Python packages downloads by major version

bigquery_pip_by_major_version.sql

-- https://bigquery.cloud.google.com/dataset/the-psf:pypi
SELECT concat(
  date(timestamp), '_',  REGEXP_EXTRACT(details.python, r'^([2-3]).[0-9].')
) as date_python, count(details.python) as downloads
FROM (TABLE_DATE_RANGE([the-psf:pypi.downloads], 
                TIMESTAMP('2016-06-26'), 
                TIMESTAMP('2016-08-31'))) 
group by date_python

bigquery_pip_by_minor_version.sql

-- https://bigquery.cloud.google.com/dataset/the-psf:pypi
-- https://bigquery.cloud.google.com/table/the-psf:pypi.downloads20160903
SELECT concat(
  date(timestamp), '_',  REGEXP_EXTRACT(details.python, r'^([2-3].[0-9]).')
) as date_python, count(details.python) as downloads
FROM (TABLE_DATE_RANGE([the-psf:pypi.downloads], 
                TIMESTAMP('2016-06-26'), 
                TIMESTAMP('2016-08-31'))) 
group by date_python

plot_python_downloads.py

#!/usr/bin/python
# -*- coding: utf-8 -*-

# To plot chart from csv generated by bigquery

import pandas as pd
import matplotlib.pyplot as plt

plt.figure()

ts = pd.read_csv('download_python_version_by_day.csv')

ts['date'] = pd.to_datetime(ts['date'])


df = ts.pivot(index='date', columns='python', values='downloads')

#df.plot()
#df[[2.6, 2.7, 3.1, 3.2, 3.3, 3.4, 3.5]].plot()
df[[2.6, 3.1, 3.2, 3.3, 3.4, 3.5]].plot()

plt.show()

plot_python_downloads_by_major_version.py

#!/usr/bin/python
# -*- coding: utf-8 -*-
import pandas as pd
import matplotlib.pyplot as plt

ts = pd.read_csv(
    'download_python_major_version_by_day.csv', parse_dates=True,
)

ts['date'] = pd.to_datetime(ts['date'])


df = ts.pivot(index='date', columns='python', values='downloads')

ax = df[[2, 3]].plot(logy=True, figsize=(12, 9))

ax.set_ylabel('log(downloads)')
ax.set_title('Python packages downloads')

plt.show()

Try a log scale on the y-axis, it's as simple as df.plot(logy=True) (docs)

Try a log scale on the y-axis, it's as simple as df.plot(logy=True) (docs)

Here's a log scale version of the same data.

Here's a log scale version of the same data.

I also think it's important for this analysis to filter down to pip installs:

SELECT
  CONCAT( DATE(timestamp), '_', REGEXP_EXTRACT(details.python, r'^([2-3].[0-9]).') ) AS date_python,
  COUNT(details.python) AS downloads
FROM (TABLE_DATE_RANGE([the-psf:pypi.downloads], TIMESTAMP('2016-06-26'), TIMESTAMP('2016-08-31')))
WHERE
  details.installer.name LIKE 'pip'
GROUP BY
  date_python

although it doesn't have a huge impact on the results.

I also think it's important for this analysis to filter down to pip installs:

SELECT
  CONCAT( DATE(timestamp), '_', REGEXP_EXTRACT(details.python, r'^([2-3].[0-9]).') ) AS date_python,
  COUNT(details.python) AS downloads
FROM (TABLE_DATE_RANGE([the-psf:pypi.downloads], TIMESTAMP('2016-06-26'), TIMESTAMP('2016-08-31')))
WHERE
  details.installer.name LIKE 'pip'
GROUP BY
  date_python

although it doesn't have a huge impact on the results.

Also, if you just want a relative comparison (rather than absolute), it might also be better to add WHERE ... AND details.cpu IS NOT NULL; more chance to get "actual" people installs rather than bots that are just mirroring (which might have more chance to be 2.7)

Also, if you just want a relative comparison (rather than absolute), it might also be better to add WHERE ... AND details.cpu IS NOT NULL; more chance to get "actual" people installs rather than bots that are just mirroring (which might have more chance to be 2.7)