Useful functions for creating visualizations
!pip3 install git+https://github.com/eddykimdata/eviz
from eviz.eviz import eviz as eviz
This function is used for calculating, for each x-axis value, the fraction contributed by each hue group. This is typically displayed as a stacked bar chart (for categorical x-values) or a line plot (for numeric or date x-values). See sections below.
The data should be a pandas dataframe, and must be in an unpivoted format. Example data:
Note that the "kind" parameter must be 'bar' or 'line', with default being 'line'.
eviz.shares_plot(
data,
x,
y,
hue,
kind='line',
cmap='Accent',
normalize=True,
group_colors=None,
barWidth=0.85,
xtick_rotation=0,
x_order=None,
hue_order=None,
marker=None,
markersize=6,
grid_height=100,
margin_plot_height=20,
margin_plot_spacer=3,
plot_totals=None,
totals_label='Total',
normalize_totals=True,
totals_marker=None,
totals_markersize=6,
height=6,
width=None,
reverse_legend_order=True)
Description of parameters:
Creates a plot (either stacked bar or line plot) describing shares of a dimension
for each given x bucket. Example: x-dimension is DS, and we want to know for
each DS, the number (or portion) of watches are from each of the 3 devices
types 'mobile', 'ctv', and 'desktop'. The input dataframe should not be
pivoted, and should be in the form:
| DS | CONTENT_CLASS | N |
|----------------:|:----------------|-----:|
| 2020-09-01 | other | 505 |
| 2020-09-01 | movie | 185 |
| 2020-09-01 | short-form | 12 |
| 2020-09-01 | series | 2135 |
| 2020-09-02 | short-form | 50 |
| 2020-09-02 | other | 713 |
Parameters
----------
data : pandas.DataFrame
Dataframe should not be pivoted.
x : str
Column name that will serve as the x ticks for the stacked bar chart
y : str
Column name that contains the values to be plotted on the y axis
hue : str
Column name that contains the groups that will be stacked
cmap : str
Name of the seaborn color map for choosing colors. Default is "Accent",
other popular cmaps include the seaborn default 'tab10', and 'Set2'.
normalize : boolean
If true, then the each stacked bar will sum to 1.
group_colors : dictionary
Dictionary map of each hue group to its color
barWidth : float
Width of each bar
xtick_rotation : float
Degrees of rotation fot he xtick labels
x_order : list
Order for the x-axis
hue_order : list
Stacking order (bottom to top) of hue groups for each bar
marker : str
Only valid for line plot, add a marker shape
markersize : int
Only valid for line plot, changes size of marker
grid_height : int
Only valid if plot_totals is not None, this number will determine the
relative height of the two subplots (main plot height + totals plot height)
margin_plot_height : int
Only valid if plot_totals is not None, this number will determine the size
of the margin plot, with the remainder of the grid_height being
allocated to the main plot's height and margin_plot_spacer.
margin_plot_spacer : int
Only valid if plot_totals is not None, this number will create additional
space between the two subplots, allocated from the total grid_height
(the main plot's height will be grid_height - margin_plot_height -
margin_plot_spacer).
plot_totals : str
Default value: None. Parameter value can be [None, 'bar', 'line'].
If not None, then a subplot will plot the totals as a bar or line plot.
totals_label : str
Label for the totals subplot legend
normalize_totals : boolean
Default value: True. If True, then the totals subplot will be normalized
such that the sum of all x-value totals will equal 1
totals_marker : str
Default: None. Only valid if plot_totals == 'line'. Adds a marker shape
to the totals subplot.
totals_markersize : int
Default: 6. Only valid if plot_totals == 'line'. Adjusts themarker size
for the totals subplot.
height : int
Default: 6. Height of the figure in inches
width : int,
Default: None. Width of the figure in inches. If None, then width
is set such that aspect ratio width:height is 4:3.
reverse_legend_order : boolean
Default: True. Changes the order of labels in the legend. If True,
top label will be the line or bar with the greatest total value (i.e.
corresponds such that top legend label will be the top bar in stacked
bar, or the top line in lineplot).
Returns
-------
fig : matplotlib.pyplot.figure
group_colors : dictionary
The mapping of each hue group to its color
Usage:
fig, group_colors = eviz.shares_plot(
data=df,
x='RELEASE_MONTH',
y='N',
x_order=None,
hue='CONTENT_CLASS',
hue_order=None,
xtick_rotation=0,
normalize=False,
cmap='tab10',
kind='bar'
)
Usage:
fig, group_colors = eviz.shares_plot(
data=df,
x='RELEASE_MONTH',
y='N',
hue='CONTENT_CLASS',
xtick_rotation=0,
normalize=True,
kind='bar',
plot_totals='bar',
normalize_totals=True,
cmap='tab10'
)
Similar to the stacked bar chart above, but the values are not stacked, and are plotted as lines.
Usage:
plt.style.use('ggplot')
fig, group_colors = eviz.shares_plot(
data=df,
x='RELEASE_MONTH',
y='N',
hue='CONTENT_CLASS',
normalize=False,
kind='line',
marker='o',
markersize=6,
)
Similar to the stacked bar chart above, but the values are not stacked, and are plotted as lines.
Usage:
fig, group_colors = eviz.shares_plot(
data=df,
x='RELEASE_MONTH',
y='N',
hue='CONTENT_CLASS',
normalize=True,
kind='line',
marker='o',
markersize=6,
plot_totals='bar',
normalize_totals=False
)
Take an existing ax and display pretty legend (text is bolded and colored, legend is placed to the right of the plot)
fig, ax = plt.subplots()
ax.plot([0,1,2], [3,4,5], label='Line A')
ax.plot([0,1,2], [6,7,9], label='Line B')
leg = eviz.get_legend(ax)
def get_dual_gridspec(
height=6,
width=8,
grid_height=100,
margin_plot_height=20,
margin_plot_spacer=3,
):
Returns a blank figure with two axes that share the x-axis.
fig, ax1, ax2 = eviz.get_dual_gridspec()
Example data. Note that the hue column "CONTENT_CLASS" must be binary (here,
the values are 'series' or 'movie').
fig, ax_top, ax_bottom = eviz.hist_mirrored(
df_runtimes,
target_col='RUNTIME_MINS',
hue='CONTENT_CLASS',
hue_order=['movie', 'series'], # optional argument
bins=np.arange(0,200,5), # Example of optional matplotlib kwargs
)
For overlay histograms, the hue can be multiclass.
fig, ax = eviz.hist_overlay(
df_runtimes,
target_col='RUNTIME_MINS',
hue='CONTENT_CLASS'
)
fig, ax = eviz.hist_overlay(
df_runtimes,
target_col='RUNTIME_MINS',
hue='CONTENT_CLASS',
bins=np.arange(0,200,5),
cumulative=True,
histtype='step',
)
def plot_histograms_by_class(
df,
dimension_cols,
hue,
output_dir=None,
width=6,
height=4,
hue_order=None,
group_colors=None,
hist_style='overlay',
bar_kwargs={},
hist_kwargs={}
):
This function will plot a histogram for all column in the given dimension_cols. For categorical columns, this function will plot a stacked barchart.
hist_style can be ['all', 'mirror', 'overlay'].