def source_vs_hour_chart(
base: alt.Chart, sensor_unit: str, max_absolute_error: float, faceted: bool = False
) -> Union[alt.Chart, alt.FacetChart]:
hd_chart = (
base.mark_rect()
.transform_joinaggregate(
on_the_fly_mae="mean(mae)",
on_the_fly_reference="mean(reference_value)",
groupby=["event_start", "source"],
)
.transform_calculate(accuracy=alt.datum.on_the_fly_mae)
.encode(
x=alt.X(
"event_start:O",
timeUnit="hours",
axis=alt.Axis(domain=False, ticks=False, labelAngle=0),
scale=alt.Scale(domain=list(range(24))),
title="Hour of day", # "UTC hour of day"
),
color=alt.condition(
selectors.time_selection_brush,
alt.Color(
"accuracy:Q",
scale=alt.Scale(
domain=(max_absolute_error, 0), scheme="redyellowgreen"
),
title="Error",
),
alt.value(selectors.idle_color),
),
tooltip=[
alt.Tooltip("event_start:T", timeUnit="hours", title="Hour of day"),
alt.Tooltip(
"accuracy:Q",
title="Mean absolute error (%s)" % sensor_unit,
format=".2f",
),
],
)
)
if faceted:
hd_chart = hd_chart.facet(
row=alt.Row("source:O", title=None, header=alt.Header(labelAngle=0))
)
else:
hd_chart = hd_chart.encode(
y=alt.Y(
"source:O",
axis=alt.Axis(domain=False, ticks=False, labelAngle=0, labelPadding=5),
title=None,
)
)
return hd_chart.properties(
title=alt.TitleParams("Model performance given a time of day", anchor="middle")
)
def confusion_matrix(df=None, truth=None, pred=None, mapping=None):
if df is None:
df = pd.DataFrame({'truth': truth, 'pred': pred})
truth = 'truth'
pred = 'pred'
threshold = len(df)
if mapping:
assert isinstance(mapping, dict), 'mapping should be a dictionary'
df[truth] = df[truth].map(lambda x: mapping[x])
df[pred] = df[pred].map(lambda x: mapping[x])
sz = 450 if len(df[truth].unique()) > 4 else 250
base = Chart(df, height=sz, width=sz).transform_aggregate(
num_vals='count()', groupby=[truth, pred]).transform_calculate(
rev_num_vals='-(datum.num_vals) + max(datum.num_vals)', ).encode(
alt.Y(f'{truth}:O', scale=alt.Scale(paddingInner=0)),
alt.X(f'{pred}:O', scale=alt.Scale(paddingInner=0)),
)
hm = base.mark_rect().encode(color=alt.Color(
'num_vals:Q', scale=alt.Scale(scheme="lightorange"), legend=None))
tx = base.mark_text(baseline='middle').encode(
text='num_vals:Q',
# color=alt.Color(alt.value('gray'))
# color='rev_num_vals:Q'
# color=alt.Color(
# 'num_vals:Q', scale=alt.Scale(scheme="redyellowgreen"),
# )
color=alt.condition(alt.datum.num_vals > threshold, alt.value('black'),
alt.value('black')))
try:
from sklearn.metrics import classification_report
print(classification_report(df[truth], df[pred]))
except:
logger.info('Skipping Report')
return hm + tx
