def test_selection_rule_picker():
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
def pick_spiky(df, train_end_time):
return [model_groups["spiky"].model_group_id]
selection_rule_picker = SelectionRulePicker(
distance_from_best_table=distance_table)
results = selection_rule_picker.results_for_rule(
bound_selection_rule=BoundSelectionRule(descriptive_name="spiky",
function=pick_spiky,
args={}),
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=["2014-01-01", "2015-01-01", "2016-01-01"],
regret_metric="precision@",
regret_parameter="100_abs",
)
assert [result["dist_from_best_case_next_time"]
for result in results] == [
0.19,
0.3,
0.12,
]
assert [result["raw_value"]
for result in results] == [0.45, 0.84, 0.45]
def test_selection_rule_picker_with_args():
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
def pick_highest_avg(df, train_end_time, metric, parameter):
assert len(df["train_end_time"].unique()) == 2
subsetted = df[(df["metric"] == metric)
& (df["parameter"] == parameter)]
mean = subsetted.groupby(["model_group_id"])["raw_value"].mean()
return [mean.nlargest(1).index[0]]
selection_rule_picker = SelectionRulePicker(
distance_from_best_table=distance_table)
regrets = [
result["dist_from_best_case_next_time"]
for result in selection_rule_picker.results_for_rule(
bound_selection_rule=BoundSelectionRule(
descriptive_name="pick_highest_avg",
function=pick_highest_avg,
args={
"metric": "recall@",
"parameter": "100_abs"
},
),
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=["2015-01-01"],
regret_metric="precision@",
regret_parameter="100_abs",
)
]
# picking the highest avg recall will pick 'spiky' for this time
assert regrets == [0.3]
def test_selection_rule_picker_with_args():
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
def pick_highest_avg(df, train_end_time, metric, parameter):
assert len(df['train_end_time'].unique()) == 2
subsetted = df[(df['metric'] == metric)
& (df['parameter'] == parameter)]
mean = subsetted.groupby(['model_group_id'])['raw_value'].mean()
return [mean.nlargest(1).index[0]]
selection_rule_picker = SelectionRulePicker(
distance_from_best_table=distance_table)
regrets = [
result['dist_from_best_case_next_time']
for result in selection_rule_picker.results_for_rule(
bound_selection_rule=BoundSelectionRule(
descriptive_name='pick_highest_avg',
function=pick_highest_avg,
args={
'metric': 'recall@',
'parameter': '100_abs'
},
),
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=['2015-01-01'],
regret_metric='precision@',
regret_parameter='100_abs',
)
]
# picking the highest avg recall will pick 'spiky' for this time
assert regrets == [0.3]
def test_SelectionRulePerformancePlotter_plot_regrets():
with patch(
"triage.component.audition.selection_rule_performance.plot_cats"
) as plot_patch:
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePerformancePlotter(
SelectionRulePicker(distance_table)
)
plotter.plot(
bound_selection_rules=[
BoundSelectionRule(
function_name="best_current_value",
args={"metric": "precision@", "parameter": "100_abs"},
),
BoundSelectionRule(
function_name="best_average_value",
args={"metric": "precision@", "parameter": "100_abs"},
),
],
regret_metric="precision@",
regret_parameter="100_abs",
model_group_ids=[1, 2],
train_end_times=["2014-01-01", "2015-01-01"],
)
assert plot_patch.called
args, kwargs = plot_patch.call_args
assert "regret" in kwargs["frame"]
assert "train_end_time" in kwargs["frame"]
assert kwargs["x_col"] == "train_end_time"
assert kwargs["y_col"] == "regret"
def test_SelectionRulePerformancePlotter_plot_metrics():
with patch('triage.component.audition.selection_rule_performance.plot_cats'
) as plot_patch:
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePerformancePlotter(
SelectionRulePicker(distance_table))
plotter.plot(bound_selection_rules=[
BoundSelectionRule(
function_name='best_current_value',
args={
'metric': 'precision@',
'parameter': '100_abs'
},
),
BoundSelectionRule(
function_name='best_average_value',
args={
'metric': 'precision@',
'parameter': '100_abs'
},
)
],
regret_metric='precision@',
regret_parameter='100_abs',
model_group_ids=[1, 2],
train_end_times=['2014-01-01', '2015-01-01'],
plot_type='metric')
assert plot_patch.called
args, kwargs = plot_patch.call_args
assert 'raw_value_next_time' in kwargs['frame']
assert 'train_end_time' in kwargs['frame']
assert kwargs['x_col'] == 'train_end_time'
assert kwargs['y_col'] == 'raw_value_next_time'
def test_SelectionPlotter_plot():
with patch("triage.component.audition.regrets.plot_cats") as plot_patch:
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePlotter(
selection_rule_picker=SelectionRulePicker(distance_table)
)
plotter.plot_all_selection_rules(
bound_selection_rules=[
BoundSelectionRule(
descriptive_name="best_current_precision",
function=best_current_value,
args={"metric": "precision@", "parameter": "100_abs"},
),
BoundSelectionRule(
descriptive_name="best_avg_precision",
function=best_average_value,
args={"metric": "precision@", "parameter": "100_abs"},
),
],
model_group_ids=[mg.model_group_id for mg in model_groups.values()],
train_end_times=["2014-01-01", "2015-01-01", "2016-01-01"],
regret_metric="precision@",
regret_parameter="100_abs",
)
assert plot_patch.called
args, kwargs = plot_patch.call_args
assert "regret" in kwargs["frame"]
assert "pct_of_time" in kwargs["frame"]
assert kwargs["x_col"] == "regret"
assert kwargs["y_col"] == "pct_of_time"
def test_SelectionPlotter_create_plot_dataframe():
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePlotter(
selection_rule_picker=SelectionRulePicker(distance_table))
plot_df = plotter.create_plot_dataframe(
bound_selection_rules=[
BoundSelectionRule(
descriptive_name="best_current_precision",
function=best_current_value,
args={
"metric": "precision@",
"parameter": "100_abs"
},
),
BoundSelectionRule(
descriptive_name="best_avg_precision",
function=best_average_value,
args={
"metric": "precision@",
"parameter": "100_abs"
},
),
],
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=["2014-01-01", "2015-01-01", "2016-01-01"],
regret_metric="precision@",
regret_parameter="100_abs",
)
# assert that we have the right # of columns and a row for each % diff value
assert plot_df.shape == (100 * 2, 3)
# both selection rules have a regret lower than 70
for value in plot_df[plot_df["regret"] == 0.70]["pct_of_time"].values:
assert np.isclose(value, 1.0)
# best avg precision rule should be within 0.14 1/3 of the time
for value in plot_df[(plot_df["regret"] == 0.14)
& (plot_df["selection_rule"] ==
"best_avg_precision")]["pct_of_time"].values:
assert np.isclose(value, 1.0 / 3)
def test_SelectionPlotter_create_plot_dataframe():
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePlotter(
selection_rule_picker=SelectionRulePicker(distance_table))
plot_df = plotter.create_plot_dataframe(
bound_selection_rules=[
BoundSelectionRule(descriptive_name='best_current_precision',
function=best_current_value,
args={
'metric': 'precision@',
'parameter': '100_abs'
}),
BoundSelectionRule(descriptive_name='best_avg_precision',
function=best_average_value,
args={
'metric': 'precision@',
'parameter': '100_abs'
}),
],
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=['2014-01-01', '2015-01-01', '2016-01-01'],
regret_metric='precision@',
regret_parameter='100_abs',
)
# assert that we have the right # of columns and a row for each % diff value
assert plot_df.shape == (100 * 2, 3)
# both selection rules have a regret lower than 70
for value in plot_df[plot_df['regret'] == 0.70]['pct_of_time'].values:
assert numpy.isclose(value, 1.0)
# best avg precision rule should be within 0.14 1/3 of the time
for value in plot_df[(plot_df['regret'] == 0.14)
& (plot_df['selection_rule'] ==
'best_avg_precision')]['pct_of_time'].values:
assert numpy.isclose(value, 1.0 / 3)
def test_SelectionPlotter_plot():
with patch('triage.component.audition.regrets.plot_cats') as plot_patch:
with testing.postgresql.Postgresql() as postgresql:
engine = create_engine(postgresql.url())
distance_table, model_groups = create_sample_distance_table(engine)
plotter = SelectionRulePlotter(
selection_rule_picker=SelectionRulePicker(distance_table))
plotter.plot_all_selection_rules(
bound_selection_rules=[
BoundSelectionRule(
descriptive_name='best_current_precision',
function=best_current_value,
args={
'metric': 'precision@',
'parameter': '100_abs'
}),
BoundSelectionRule(descriptive_name='best_avg_precision',
function=best_average_value,
args={
'metric': 'precision@',
'parameter': '100_abs'
}),
],
model_group_ids=[
mg.model_group_id for mg in model_groups.values()
],
train_end_times=['2014-01-01', '2015-01-01', '2016-01-01'],
regret_metric='precision@',
regret_parameter='100_abs',
)
assert plot_patch.called
args, kwargs = plot_patch.call_args
assert 'regret' in kwargs['frame']
assert 'pct_of_time' in kwargs['frame']
assert kwargs['x_col'] == 'regret'
assert kwargs['y_col'] == 'pct_of_time'
