Upgrade Progression

Upgrade Progression.ipynb

Upgrade Progression.py

# coding: utf-8

# ### Longitudinal Dataset Tutorial

# In[1]:

import pandas as pd
import numpy as np
import plotly.plotly as py

get_ipython().magic(u'pylab inline')


# In[2]:

sc.defaultParallelism


# In[4]:

frame = sqlContext.sql("SELECT subsession_start_date, build, settings FROM longitudinal")


# Number of profiles:

# In[5]:

frame.count()


# In[8]:

few = frame.rdd.take(10)


# In[29]:

get_ipython().run_cell_magic(u'time', u'', u"\nfrom datetime import datetime, date\nfrom itertools import imap\n\ntotal_records = sc.accumulator(0)\nnone_records = sc.accumulator(0)\nnon_release = sc.accumulator(0)\nvalue_error = sc.accumulator(0)\n\ndef map_upgrades(row):\n    total_records.add(1)\n    \n    if row.settings is None or row.subsession_start_date is None or row.build is None:\n        none_records.add(1)\n        return\n    \n    # ignore people who aren't release-only or have really weird data\n    for s in row.settings:\n        if s.update.channel != 'release':\n            non_release.add(1)\n            return\n\n    try:\n        versions = [tuple(imap(int, b.version.split('.'))) for b in row.build]\n        dates = [datetime.strptime(ssd[:10], '%Y-%m-%d').date()\n                 for ssd in row.subsession_start_date]\n    except (ValueError, AttributeError):\n        value_error.add(1)\n        return\n    \n    version_dates = zip(versions, dates)\n    version_dates.reverse()\n    \n    # If the first session we see isn't running up against the 6-month window\n    # don't record the first session as an upgrade\n    \n    last_version = None\n    version, ssd = version_dates[0]\n    if ssd < date(2016, 1, 1):\n        last_version = version\n        \n    for version, ssd in version_dates:\n        if version > last_version:\n            yield ssd, version\n            last_version = version\n            \nr = frame.flatMap(map_upgrades).countByValue()")


# In[33]:

print total_records.value
print none_records.value
print non_release.value
print value_error.value


# In[44]:

from datetime import date, datetime
updates_46_0_1 = [(d, c) for ((d, v), c) in r.iteritems()
                  if v == (45, 0, 1)
                    and d > date(2016, 3, 1) and d < date(2016, 5, 11)]
updates_46_0_1.sort(key=lambda (d, c): d)


# In[45]:

import plotly.plotly as py
import plotly.graph_objs as go

d = [
    go.Bar(
        x=[d for (d, c) in updates_46_0_1],
        y=[c for (d, c) in updates_46_0_1]
    )
]
py.iplot(d)


# In[71]:

idx = pd.Series()
f = pd.DataFrame((c for d, c in updates_46_0_1), columns=["c"], index=(pd.to_datetime(d) for d, c in updates_46_0_1))
fig = plt.figure(figsize=(11, 4))
plt.title("Upgrades to 45.0.1 over time")
plt.ylabel("Upgrades per day")
plt.xlabel("Date of First 45.0.1 run")
plt.plot(f.index, f)
plt.show()