mybuddy/reports/graphs.py

215 lines
7.1 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
from __future__ import unicode_literals
2017-08-25 18:17:15 +00:00
from collections import OrderedDict
from django.db.models import Count, Case, When
from django.db.models.functions import TruncDate
from django.utils import timezone
import pandas as pd
import plotly.offline as plotly
import plotly.graph_objs as go
from core.models import DiaperChange, Sleep
from core.utils import duration_string
from .utils import default_graph_layout_options, split_graph_output
def diaperchange_types(child):
changes = DiaperChange.objects.filter(child=child) \
.annotate(date=TruncDate('time')) \
.values('date') \
.annotate(wet_count=Count(Case(When(wet=True, then=1)))) \
.annotate(solid_count=Count(Case(When(solid=True, then=1)))) \
.annotate(total=Count('id')) \
.order_by('-date')
solid_trace = go.Scatter(
mode='markers',
name='Solid changes',
x=list(changes.values_list('date', flat=True)),
y=list(changes.values_list('solid_count', flat=True)),
)
wet_trace = go.Scatter(
mode='markers',
name='Wet changes',
x=list(changes.values_list('date', flat=True)),
y=list(changes.values_list('wet_count', flat=True))
)
total_trace = go.Scatter(
name='Total changes',
x=list(changes.values_list('date', flat=True)),
y=list(changes.values_list('total', flat=True))
)
layout_args = default_graph_layout_options()
layout_args['barmode'] = 'stack'
2017-09-13 15:20:28 +00:00
layout_args['title'] = '<b>Diaper Change Types</b><br>{}'.format(child)
layout_args['xaxis']['title'] = 'Date'
layout_args['yaxis']['title'] = 'Number of changes'
fig = go.Figure({
'data': [solid_trace, wet_trace, total_trace],
'layout': go.Layout(**layout_args)
})
output = plotly.plot(fig, output_type='div', include_plotlyjs=False)
return split_graph_output(output)
2017-09-10 21:26:18 +00:00
def sleep_amount(child):
"""Create a graph showing total time sleeping for each day."""
instances = Sleep.objects.filter(child=child).order_by('start')
totals = {}
for instance in instances:
start_time = timezone.localtime(instance.start)
start_date = start_time.date().isoformat()
if start_date not in totals.keys():
totals[start_date] = timezone.timedelta()
totals[start_date] += instance.duration
def sleep_times(child):
"""Create a graph showing blocked out periods of sleep during each day."""
instances = Sleep.objects.filter(child=child).order_by('start')
2017-08-25 18:17:15 +00:00
y_df = pd.DataFrame()
text_df = pd.DataFrame()
last_end_time = None
adjustment = None
df_index = 0
for instance in instances:
start_time = timezone.localtime(instance.start)
end_time = timezone.localtime(instance.end)
start_date = start_time.date().isoformat()
duration = instance.duration
# Check if the previous entry crossed midnight (see below).
if adjustment:
# Fake (0) entry to keep the color switching logic working.
df_index = _add_sleep_entry(
y_df, text_df, 0, adjustment['column'], 0)
# Real adjustment entry.
df_index = _add_sleep_entry(
y_df,
text_df,
df_index,
adjustment['column'],
adjustment['duration'].seconds/60,
'Asleep {} ({} to {})'.format(
duration_string(adjustment['duration']),
adjustment['start_time'].strftime('%I:%M %p'),
adjustment['end_time'].strftime('%I:%M %p')
)
)
last_end_time = timezone.localtime(adjustment['end_time'])
adjustment = None
# If the dates do not match, set up an adjustment for the next day.
if end_time.date() != start_time.date():
adj_start_time = end_time.replace(hour=0, minute=0, second=0)
adjustment = {
'column': end_time.date().isoformat(),
'start_time': adj_start_time,
'end_time': end_time,
'duration': end_time - adj_start_time
}
# Adjust end_time for the current entry.
end_time = end_time.replace(
day=start_time.day, hour=23, minute=59, second=0)
duration = end_time - start_time
2017-08-25 18:17:15 +00:00
if start_date not in y_df:
last_end_time = start_time.replace(hour=0, minute=0, second=0)
2017-08-25 18:17:15 +00:00
# Awake time.
df_index = _add_sleep_entry(
y_df,
text_df,
2017-08-25 18:17:15 +00:00
df_index,
start_date,
(start_time - last_end_time).seconds/60
2017-08-25 18:17:15 +00:00
)
# Asleep time.
df_index = _add_sleep_entry(
y_df,
text_df,
2017-08-25 18:17:15 +00:00
df_index,
start_date,
duration.seconds/60,
'Asleep {} ({} to {})'.format(
duration_string(duration),
2017-08-25 18:17:15 +00:00
start_time.strftime('%I:%M %p'),
end_time.strftime('%I:%M %p')
2017-08-25 18:17:15 +00:00
)
)
last_end_time = end_time
2017-08-25 18:17:15 +00:00
dates = list(y_df)
traces = []
color = 'rgba(255, 255, 255, 0)'
2017-08-25 18:17:15 +00:00
for index, row in y_df.iterrows():
traces.append(go.Bar(
x=dates,
y=row,
2017-08-25 18:17:15 +00:00
text=text_df.ix[index],
hoverinfo='text',
marker={'color': color},
showlegend=False,
))
if color == 'rgba(255, 255, 255, 0)':
color = 'rgb(0, 0, 255)'
else:
color = 'rgba(255, 255, 255, 0)'
layout_args = default_graph_layout_options()
2017-08-25 18:17:15 +00:00
layout_args['margin']['b'] = 100
layout_args['barmode'] = 'stack'
layout_args['hovermode'] = 'closest'
2017-09-13 15:20:28 +00:00
layout_args['title'] = '<b>Sleep Patterns</b><br>{}'.format(child)
2017-08-25 18:17:15 +00:00
layout_args['height'] = 600
layout_args['xaxis']['title'] = 'Date'
layout_args['xaxis']['type'] = 'category'
2017-08-25 18:17:15 +00:00
layout_args['xaxis']['tickangle'] = -65
start = timezone.localtime().strptime('12:00 AM', '%I:%M %p')
ticks = OrderedDict()
ticks[0] = start.strftime('%I:%M %p')
for i in range(30, 60*24, 30):
ticks[i] = (start + timezone.timedelta(minutes=i)).strftime('%I:%M %p')
layout_args['yaxis']['title'] = 'Time of day'
layout_args['yaxis']['rangemode'] = 'tozero'
layout_args['yaxis']['tickmode'] = 'array'
2017-08-25 18:17:15 +00:00
layout_args['yaxis']['tickvals'] = list(ticks.keys())
layout_args['yaxis']['ticktext'] = list(ticks.values())
layout_args['yaxis']['tickfont'] = {'size': 10}
fig = go.Figure({
'data': traces,
'layout': go.Layout(**layout_args)
})
output = plotly.plot(fig, output_type='div', include_plotlyjs=False)
return split_graph_output(output)
def _add_sleep_entry(y_df, text_df, index, column, duration, text=''):
"""Create a duration and text description entry in a DataFrame and return
the next index on success."""
if column not in y_df:
y_df.assign(**{column: 0 in range(0, len(y_df.index))})
text_df.assign(**{column: 0 in range(0, len(text_df.index))})
index = 0
y_df.set_value(index, column, duration)
text_df.set_value(index, column, text)
return index + 1