def test_plot_bar_returns_a_bar_plot_with_correct_elements(self):
plot_data = TESTDATA_1
x_var = 'snapshot_date'
y_var = 'pos_count'
title = '# positive examples over time'
subplot_index = 0
_, axes = plt.subplots(nrows=2, ncols=1)
viz_utils.plot_bar(plot_data=plot_data,
x_variable=x_var,
y_variable=y_var,
title=title,
axes=axes,
subplot_index=subplot_index)
bar_plot = axes[subplot_index]
x_data = list(plot_data[x_var])
y_data = [float(y) for y in list(plot_data[y_var])]
with self.subTest(name='test x axis variable is equal'):
self.assertEqual(x_var, bar_plot.get_xlabel())
with self.subTest(name='test x axis data is equal'):
self.assertListEqual(x_data, [
tick.get_text()
for tick in bar_plot.get_xticklabels(which='major')
])
with self.subTest(name='test y axis variable is equal'):
self.assertEqual(y_var, bar_plot.get_ylabel())
with self.subTest(name='test y axis data is equal'):
self.assertListEqual(y_data,
[h.get_height() for h in bar_plot.patches])
with self.subTest(name='test title is equal'):
self.assertEqual(title, bar_plot.get_title())
def _plot_categorical_fact(
plot_data: pd.DataFrame, fact_name: str,
plot_style_params: _FactPlotStyles) -> List[axes.Axes]:
"""Plots the statistics of a categorical fact variable.
Generates plots of daily record count, latest distribution of top N levels
and daily distribution of top N level over time of a categorical fact
variable.
Args:
plot_data: Data to plot containing date, total_count, value and percentage
columns.
fact_name: Name of the fact variable.
plot_style_params: Plot style parameters.
Returns:
plots: A list of Axes containing 3 plots.
"""
logging.info('Plotting categorical fact %s ', fact_name)
_, plots = pyplot.subplots(nrows=_ROWS_IN_SUBPLOTS_GRID,
ncols=_COLS_IN_SUBPLOTS_GRID,
figsize=(plot_style_params.fig_width,
plot_style_params.fig_height))
latest_date = max(plot_data['date'])
latest_date_stats = plot_data[plot_data['date'] ==
latest_date].sort_values(by=['percentage'],
ascending=False)
common_lineplot_params = {
'axes': plots,
'title_fontsize': plot_style_params.lineplot_title_fontsize,
'xticklabels_fontsize':
plot_style_params.lineplot_xticklabels_fontsize,
'yticklabels_fontsize': plot_style_params.lineplot_yticklabels_fontsize
}
# plot daily total fact count
viz_utils.plot_line(plot_data=plot_data[['date', 'total_record_count'
]].drop_duplicates(),
x_variable='date',
y_variable='total_record_count',
title=f'{fact_name} - Daily Fact Count',
subplot_index=0,
line_color=plot_style_params.line_color_record_count,
**common_lineplot_params)
# plot the latest distribution of the top N fact levels.
viz_utils.plot_bar(
plot_data=latest_date_stats,
x_variable='category_value',
y_variable='percentage',
title=f'{fact_name} - Latest Value Distribution (%)',
axes=plots,
subplot_index=1,
title_fontsize=plot_style_params.barplot_title_fontsize,
xticklabels_fontsize=plot_style_params.barplot_xticklabels_fontsize,
yticklabels_fontsize=plot_style_params.barplot_yticklabels_fontsize)
# plot the daily distribution of the top N fact levels over time.
viz_utils.plot_line(
plot_data=plot_data,
x_variable='date',
y_variable='percentage',
title=f'{fact_name} - Daily value distribution (%)',
subplot_index=2,
category_variable='category_value',
legend_fontsize=plot_style_params.lineplot_legend_fontsize,
**common_lineplot_params)
return plots
def _plot_categorical_feature_numerical_label(
label_stats_data: pd.DataFrame, feature_stats_data: pd.DataFrame,
feature_name: str,
plot_style_params: _FeaturePlotStyles) -> List[axes.Axes]:
"""Plots the statistics of a categorical feature when label is numerical.
Generates following plots:
- distribution of the label (box plots) for different category values of the
feature.
- distribution of feature values (stacked bor plots) by snapshot_date.
Args:
label_stats_data: Plot data containing the following columns: feature,
value, mean, stddev, med, q1, q3, whislo and whishi.
feature_stats_data: Plot data containing the following columns:
snapshot_date, record_count, prop_missing, prop_non_num, average and
stddev.
feature_name: Name of the feature.
plot_style_params: Plot style parameters.
Returns:
plots: A list of Axes containing 2 plots.
"""
logging.info('Plotting categorical feature %s', feature_name)
_, plots = pyplot.subplots(nrows=_NO_PLOTS_CAT_FEATURE_NUM_LABEL,
ncols=_COLS_IN_SUBPLOTS_GRID,
figsize=(plot_style_params.fig_width,
plot_style_params.fig_height))
# Plot distribution of the label by different feature values (categories)
logging.info('Plotting label distribution by feature category values.')
viz_utils.plot_box(
plot_data=label_stats_data,
title=f'Label distribution by [{feature_name}] categories',
axes=plots,
subplot_index=0,
x_variable='value',
x_label='Category value',
y_label='Label distribution',
title_fontsize=plot_style_params.title_fontsize,
xlabel_fontsize=plot_style_params.xlabel_fontsize,
ylabel_fontsize=plot_style_params.ylabel_fontsize,
xticklabels_fontsize=plot_style_params.xticklabels_fontsize,
yticklabels_fontsize=plot_style_params.yticklabels_fontsize,
xticklabels_rotation=0)
# Plot snapshot-level feature distribution
viz_utils.plot_bar(
plot_data=feature_stats_data,
x_variable='snapshot_date',
y_variable='percentage',
group_variable='value',
stacked_bars=True,
title=f'Snapshot-level distribution of [{feature_name}]',
subplot_index=1,
axes=plots,
title_fontsize=plot_style_params.title_fontsize,
xlabel_fontsize=plot_style_params.xlabel_fontsize,
ylabel_fontsize=plot_style_params.ylabel_fontsize,
xticklabels_fontsize=plot_style_params.xticklabels_fontsize,
yticklabels_fontsize=plot_style_params.yticklabels_fontsize,
xticklabels_rotation=45,
)
return plots
def _plot_categorical_feature_binary_label(
df_data: pd.DataFrame, feature_name: str, label_column: str,
positive_class_label: LabelType, negative_class_label: LabelType,
plot_style_params: _FeaturePlotStyles) -> List[axes.Axes]:
"""Plots the statistics of a categorical feature when label is binary.
Generates following plots of the feature:
- distribution of values (bor plots) by label.
- distribution of values (stacked bor plots) for positive instances by
snapshot_date.
- distribution of values (stacked bor plots) for negative instances by
snapshot_date.
Args:
df_data: plot data containing the following columns: snapshot_date, label,
record_count, prop_missing, prop_non_num, average, stddev columns.
feature_name: Name of the feature.
label_column: Name of the label column.
positive_class_label: label for positive class
negative_class_label: label for negative class
plot_style_params: Plot style parameters.
Returns:
plots: A list of Axes containing 3 plots.
"""
logging.info('Plotting categorical feature %s', feature_name)
_, plots = pyplot.subplots(nrows=_NO_PLOTS_CAT_FEATURE_BINARY_LABEL,
ncols=_COLS_IN_SUBPLOTS_GRID,
figsize=(plot_style_params.fig_width,
plot_style_params.fig_height))
# Aggregating dataframe on date level to get data for the category
# distribution plot.
df_value_count = df_data.groupby([label_column,
'value'])[['count']].sum().reset_index()
df_total_count = df_data.groupby(label_column)[['count'
]].sum().reset_index()
df_total_count = df_total_count.rename(columns={'count': 'total_count'})
# Joining total counts and calculating proportions.
df_value_proportions = df_value_count.merge(df_total_count,
on=label_column)
df_value_proportions['percentage'] = (
df_value_proportions['count'] /
df_value_proportions['total_count']) * 100
common_barplot_params = {
'axes': plots,
'title_fontsize': plot_style_params.title_fontsize,
'xlabel_fontsize': plot_style_params.xlabel_fontsize,
'ylabel_fontsize': plot_style_params.ylabel_fontsize,
'xticklabels_fontsize': plot_style_params.xticklabels_fontsize,
'yticklabels_fontsize': plot_style_params.yticklabels_fontsize
}
# Plot distribution of the feature values by label
viz_utils.plot_bar(plot_data=df_value_proportions,
x_variable='value',
y_variable='percentage',
group_variable=label_column,
title=f'Distribution of [{feature_name}]',
subplot_index=0,
**common_barplot_params)
# Plot the snapshot-level distribution of the feature for positive instances
pos_instance_stats = df_data[df_data[label_column] == positive_class_label]
pos_instance_stats = pos_instance_stats.sort_values(
['snapshot_date', 'feature'], ascending=True)
pos_plot_title = (f'Snapshot-level distribution of [{feature_name}] for '
f'label = {positive_class_label}')
viz_utils.plot_bar(
plot_data=pos_instance_stats,
x_variable='snapshot_date',
y_variable='percentage',
group_variable='value',
stacked_bars=True,
title=pos_plot_title,
subplot_index=1,
xticklabels_rotation=45,
x_label='', # to better arrange the output plots
**common_barplot_params)
# Plot the snapshot-level distribution of the feature for negative instances
neg_instance_stats = df_data[df_data[label_column] == negative_class_label]
neg_instance_stats = neg_instance_stats.sort_values(
['snapshot_date', 'feature'], ascending=True)
neg_plot_title = (f'Snapshot-level distribution of [{feature_name}] for '
f'label = {negative_class_label}')
viz_utils.plot_bar(
plot_data=neg_instance_stats,
x_variable='snapshot_date',
y_variable='percentage',
group_variable='value',
stacked_bars=True,
title=neg_plot_title,
subplot_index=2,
xticklabels_rotation=45,
**common_barplot_params,
)
return plots