## Saturday, December 29, 2018

### Apple Health with Python

For about four years I have been storing my weight in the Apple Health app. At the same time, I am almost always carrying my phone so it has been recording my number of steps for at least as long. As Python seems to be a hot topic at the moment, I thought I would try using it to look for correlations between my weight and number of steps.

The whole project would have been way easier in Excel, but it seemed a good way to learn plotting and mathematics in Python.

First I plotted my weight and number of steps:
The weight plot definitely supports the hypothesis that it is easy to lose weight in the short term, and almost impossible over a five year period. I managed to drop fast when I first started recording but then slowly gained it all back and then some. The steps are more opaque. At first glance they don't say much.

So I tried plotting steps versus weight for the days where I have both sets of data. The X axis is how many steps I took that day, and the Y axis is what I weighed that morning:
Better, this shows that when I am above 190lbs I rarely walk more than 10,000 steps, but it is still an amorphous blob with a whole bunch of noise.

Some of the noise can be removed by instead plotting monthly averages. For every 30 day period in the dataset I averaged my steps and averaged by weight:
Again an amorphous blob, but  it does show that when I walk a lot I am more likely to be low weight and when I am heavy I am more likely to not walk much.

That seemed like it missed an important detail though, am I 190 pounds and dropping fast because I walk so much? or am I 180 pounds and gaining weight fast? That problem can be removed by instead looking at weight gain. So I averaged my number of steps for a 30 day period, then compared my weight in that 30 day period to my weight in the previous 30 day period:

There is still a lot of noise, but when I lose more than two pounds in a month I always am walking more than 6,500 steps, and when I gain more than a pound in a week I am always walking less than 7,000 steps.

The code to do this simple analysis started with trying to get the data out of the export.xml file which apple provides. This was easy enough in excel, but I made it a point to use python. Eventually I came up with the following script which pulls all the weight and steps from the file:
import xml.etree.ElementTree
# This file location will need to be edited to match where the file is.
xDoc = xml.etree.ElementTree.parse(
'E:/My Documents/Python files/export/apple_health_data/export.xml')

items = list(xDoc.getroot()) # Convert the XML items to a list

step_data = []

# Searches for steps in the XML file.
item_type_identifier='HKQuantityTypeIdentifierStepCount' # Desired data type
for i,item in enumerate(items):
if 'type' in item.attrib and item.attrib['type'] == item_type_identifier:
# Attributes to extract from the current item
step_data.append((item.attrib['type'],
item.attrib['endDate'],
item.attrib['value']))

weight_data = []
# Searches for weights in the XML file
item_type_identifier='HKQuantityTypeIdentifierBodyMass'
for i,item in enumerate(items):
if 'type' in item.attrib and item.attrib['type'] == item_type_identifier:
# Attributes to extract from the current item
weight_data.append((item.attrib['type'],
item.attrib['endDate'],
item.attrib['value']))

file_location = \
'E:/My Documents/Python files/export/apple_health_data/Extract.csv'
file = open(file_location, 'w')

# Writes the list to a csv file by putting a , after every line and a \n at the
# end of a row.
i= 0
for i in range(len(step_data)):
file.write(str(step_data[i])+',')
file.write(str(step_data[i])+',')
file.write(str(step_data[i])+',')
file.write(str('\n'))
i = i + 1

i= 0
for i in range(len(weight_data)):
file.write(str(weight_data[i])+',')
file.write(str(weight_data[i])+',')
file.write(str(weight_data[i])+',')
file.write(str('\n'))
i = i + 1

file.close()
I could have just used the data directly, but I chose to divide this up into three scripts. Now that I have all the data in a csv file it is necessary to do some basic analysis on it to make the plots. The averaging and summing necessary was done in a second script which imports the data from the previous script and again saves to csv:

import csv
import pandas as pd
import datetime

file_location = \
'E:/My Documents/Python files/export/apple_health_data/Extract.csv'

weights = []
weight = []

# This pulls the weights data out of the csv file and puts it into the object
# called weights
with open(file_location) as csvfile:

if row == 'HKQuantityTypeIdentifierBodyMass':

pounds = float(row)

date_string = row.split()
year_int = int(date_string.split('-'))
month_int = int(date_string.split('-'))
day_int = int(date_string.split('-'))

date = datetime.date(year_int, month_int, day_int)

weight = [date, pounds]
weights.append(weight)

step_counts = []
step_count = []
# This pulls the steps data out of the csv file and puts it into the object
# called step_counts
with open(file_location) as csvfile:

if row == 'HKQuantityTypeIdentifierStepCount':

steps = int(row)
date_string = row.split()
year_int = int(date_string.split('-'))
month_int = int(date_string.split('-'))
day_int = int(date_string.split('-'))
date = datetime.date(year_int, month_int, day_int)

step_count = [date, steps]
step_counts.append(step_count)

# The steps from Apple come in smaller batches than make sense for this analysis
# Therefore I added up all the steps on any particular day and put it in the
# object daily_steps
summed_steps = []
daily_steps = 0
for entry in step_counts:
if entry == date:
daily_steps += int(entry)
else:
days_steps = [date, daily_steps]
summed_steps.append(days_steps)
date = entry
daily_steps = int(entry)

# The weights and steps were moved from a list to a dataframe for ease of some
# of the later anaylsis
weights_data_frame=pd.DataFrame(weights,columns=['date', 'weight'])
steps_data_frame = pd.DataFrame(summed_steps,columns=['date','steps'])

# The weights and steps were combined into one dataframe.
combined_data = steps_data_frame.merge(weights_data_frame, how='outer')

# In case any data is out of order it is sorted
combined_data = combined_data.sort_values(by=['date'])

# Calculates the 7 day average weight.
weekly_average_weight = combined_data["weight"].rolling(min_periods=1,
center=True,
window=7).mean()
weekly_average_weight = weekly_average_weight.to_frame('weekly average weight')
combined_data = combined_data.join(weekly_average_weight)

# Calculates the 7 day average steps.
weekly_average_steps = combined_data["steps"].rolling(min_periods=1,
center=True,
window=7).mean()
weekly_average_steps = weekly_average_steps.to_frame('weekly average steps')
combined_data = combined_data.join(weekly_average_steps)

# Calculates the 30 day average weight.
monthly_average_weight = combined_data["weight"].rolling(min_periods=1,
center=True,
window=30).mean()
monthly_average_weight = monthly_average_weight.to_frame(
'monthly average weight')
combined_data = combined_data.join(monthly_average_weight)

# Calculates the 30 day average steps.
monthly_average_steps = combined_data["steps"].rolling(min_periods=1,
center=True,
window=30).mean()
monthly_average_steps = monthly_average_steps.to_frame('monthly average steps')
combined_data = combined_data.join(monthly_average_steps)

# Compares the average weight to the average weight the week before to look
#for a gain.
i = 0
weekly_gain = []
for i in range(len(combined_data)-4):
if combined_data.iloc[i+4] and combined_data.iloc[i-3] \
and i +4< len(combined_data) and i-4>=0:
gain = combined_data.iloc[i+4] - combined_data.iloc[i-3]

dated_gain = [combined_data.iloc[i], gain]
weekly_gain.append(dated_gain)
i += 1

# Puts the gain into a dataframe
weekly_gain = pd.DataFrame(weekly_gain,columns=['date','weekly weight gain'])

# Merges the dataframe
combined_data = combined_data.merge(weekly_gain, how='outer')

# Compares the average weight to the average weight the month before to look
# for a gain.
i = 0
monthly_gain = []
for i in range(len(combined_data)-30):
if combined_data.iloc[i+30] and combined_data.iloc[i-30] \
and i +15< len(combined_data) and i-30>=0:
gain = combined_data.iloc[i] - combined_data.iloc[i-30]
print(combined_data.iloc[i])
date_1 = combined_data.iloc[i]

dated_gain = [date_1, gain]
print(dated_gain)
monthly_gain.append(dated_gain)
i += 1
# Puts the gain into a dataframe
monthly_gain = pd.DataFrame(monthly_gain,columns=['date','monthly weight gain'])
print(monthly_gain)
# Merges the dataframe
combined_data = combined_data.merge(monthly_gain, how='outer')

print(combined_data)

combined_data.to_csv("output.csv")
Finally I had to make a bunch of plots out of the data which was stored to the csv. This was done in a third script which imported the formatted data from the second script.

import matplotlib
import matplotlib.pyplot as plt
import pandas as pd

# Imports the csv, will need to change file path to match the location.
file_path = 'E:/My Documents/Python files/export/apple_health_data/output.csv'

# Puts the imported data into a dataframe
combined_data['date'] = pd.to_datetime(combined_data['date'])

# Start of a plot of date versus monthly average steps, first line defines
# which values are plotted against each other.

plt.plot_date(combined_data['date'], combined_data['monthly average steps'])

# Rotates the labels since dates get long otherwise.
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)

# Labels the y axis, x axis not named since dates seem obvious.
plt.ylabel('Average Steps in 30 Day Period')

# The plot is made larger then saved.
fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('monthly_average_steps.png', dpi=100)

# Marks the end of one plot.
plt.figure()

# Start of second plot.
plt.plot_date(combined_data['date'], combined_data['monthly average weight'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)

plt.ylabel('Average Weight in 30 Day Period')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('monthly_average_weight.png', dpi=100)

plt.figure()

# Start of third plot, this one is a bit different since it is a scatterplot
# rather than a date plot.
plt.scatter(combined_data['monthly average steps'], combined_data['weight'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)
plt.xlabel('Average Steps in 30 Day Period')
plt.ylabel('weight')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('monthly_average_steps_versus_weight.png', dpi=100)

plt.figure()

# Start of fourth plot
plt.scatter(combined_data['monthly average steps'], combined_data['monthly weight gain'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)
plt.xlabel('Average Steps in 30 Day Period')
plt.ylabel('Weight Gain Compared to Previous 30 Day Period')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('monthly_average_steps_versus_weight_gain.png', dpi=100)

plt.figure()

# Start of fifth plot
plt.scatter(combined_data['monthly average steps'], combined_data['monthly average weight'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)
plt.xlabel('Average Steps in 30 Day Period')
plt.ylabel('Monthly Average Weight')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('monthly_average_steps_versus_monthly_average_weight.png', dpi=100)

plt.figure()

# Start of sixth plot
plt.plot_date(combined_data['date'], combined_data['steps'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)

plt.ylabel('daily steps')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('steps.png', dpi=100)

plt.figure()

# Start of seventh plot
plt.plot_date(combined_data['date'], combined_data['weight'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)

plt.ylabel('daily weight')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('weight.png', dpi=100)

plt.figure()

# Start of eight plot
plt.scatter(combined_data['steps'], combined_data['weight'])
plt.setp(plt.gca().xaxis.get_majorticklabels(),rotation=90,)
plt.xlabel('Steps')
plt.ylabel('Weight')

fig = matplotlib.pyplot.gcf()
fig.set_size_inches(12, 8)
fig.savefig('steps_versus_weight.png', dpi=100)

plt.figure()