Skip to content

Commit

Permalink
Add support for Python > 3.8 in regressions package
Browse files Browse the repository at this point in the history
  • Loading branch information
@NPodlozhniy
NPodlozhniy committed Aug 11, 2024
1 parent 72c6fba commit b6c5230
Show file tree
Hide file tree
Showing 2 changed files with 66 additions and 35 deletions.
99 changes: 65 additions & 34 deletions podlozhnyy_module/regression.py
View file
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import sys
from functools import wraps
from itertools import product
from math import log
Expand All @@ -20,7 +21,11 @@ def wrapper(*args, **kwargs):
diagramm = func(*args, **kwargs)
for bnd, opts in [("matplotlib", matplotlib_opts), ("bokeh", bokeh_opts)]:
if bnd in hv.Store._options and bnd == hv.Store.current_backend:
return diagramm.opts(opts)
return (
hv.opts.apply_groups(diagramm, opts)
if sys.version_info > (3, 8)
else diagramm.opts(opts)
)
return diagramm

return wrapper
Expand Down Expand Up @@ -63,13 +68,22 @@ def wrapper(*args, **kwargs):
"plot": dict(show_grid=True, xrotation=45),
"style": dict(facecolor=colors, alpha=0.3),
},
"Bars.Observed_Bad_Rate": {
"plot": dict(show_legend=True, xrotation=90),
"style": dict(facecolor=colors.values[0]),
},
"Curve.Predicted_Bad_Rate": {
"plot": dict(show_legend=True, xrotation=90),
"style": dict(facecolor=colors.values[1]),
},
"Overlay.Gain_Chart": {
"plot": dict(legend_position="top_left"),
},
}

bokeh_opts = {
"Scatter.Weight_of_Evidence": {
"plot": dict(
show_grid=True, tools=["hover"], legend_position="right", width=450
),
"plot": dict(show_grid=True, tools=["hover"], legend_position="right", width=450),
"style": dict(color="r", size=5),
},
"NdOverlay.Objects_rate": {
Expand All @@ -90,10 +104,21 @@ def wrapper(*args, **kwargs):
"plot": dict(show_grid=True, xrotation=45),
"style": dict(color=colors, alpha=0.3),
},
"Bars.Observed_Bad_Rate": {
"plot": dict(show_legend=True, xrotation=90),
"style": dict(color=colors.values[0]),
},
"Curve.Predicted_Bad_Rate": {
"plot": dict(show_legend=True, xrotation=90),
"style": dict(color=colors.values[1]),
},
"Overlay.Gain_Chart": {
"plot": dict(legend_position="top_left"),
},
}


def make_bucket(df, feature: str, num_buck: int=10, group_size: str=None):
def make_bucket(df, feature: str, num_buck: int = 10, group_size: str = None):
"""
Производит разбиение на бакеты
Expand All @@ -118,18 +143,17 @@ def make_bucket(df, feature: str, num_buck: int=10, group_size: str=None):
.cumsum(skipna=True)
)
totals = (
data_ranked
.groupby(feature)
data_ranked.groupby(feature)
.agg({group_size: "max"})
.rename(columns={group_size: "total"})
)
bucket = np.ceil(
data_ranked.join(totals, on=feature)["total"] / totals["total"].max() * num_buck
data_ranked.join(totals, on=feature)["total"]
/ totals["total"].max()
* num_buck
).fillna(num_buck + 1)
else:
bucket = np.ceil(
df[feature].rank(pct=True) * num_buck
).fillna(num_buck + 1)
bucket = np.ceil(df[feature].rank(pct=True) * num_buck).fillna(num_buck + 1)

agg = df[feature].groupby(bucket).agg(["min", "max"])

Expand Down Expand Up @@ -201,15 +225,16 @@ def check_linearity(df, feature, target, num_buck=10):
num_buck: Количество бакетов, если признак числовой
"""
return (
df.pipe(make_bucket, feature, num_buck)
df[[feature, target]]
.pipe(make_bucket, feature, num_buck)
.groupby("bucket")
.mean()
.pipe(
lambda x: hv.Scatter(
zip(np.array(x[feature]), np.array(x[target])),
kdims=f"{feature}",
vdims=f"{target}",
label=f"Проверка линейности зависимости {target} от {feature}",
label=f"Checking the linearity of dependence {target} on {feature}",
)
* simple_reg(np.array(x[feature]), np.array(x[target]))
)
Expand Down Expand Up @@ -238,7 +263,7 @@ def get_residuals(y, pred):
zip(predicts, get_residuals(df[target], predicts)),
kdims=["Estimated target"],
vdims=["Residual"],
label=f"Проверка гомоскедастичности признака {feature}",
label=f"Homoscedasticity check of {feature}",
)


Expand Down Expand Up @@ -280,7 +305,7 @@ def _woe_confint(n, cnt, q):
return _woe(p_low, q), _woe(p_high, q)


def bad_rate(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
def bad_rate(df, feature: str, target: str, num_buck: int = 10, group_size: str = None):
"""
Считает bad_rate для каждого бакета признака в модели классификации.
Возвращает датафрейм с аггрегациями (сумма таргета в бакете, среднее значение предсказания,
Expand Down Expand Up @@ -316,7 +341,7 @@ def bad_rate(df, feature: str, target: str, num_buck: int=10, group_size: str=No
)


def woe(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
def woe(df, feature: str, target: str, num_buck: int = 10, group_size: str = None):
"""
Считает WOE для признака в модели классификации.
Доля объектов каждого класса ограничивается 0.001 - снизу и 0.999 - сверху.
Expand All @@ -328,9 +353,9 @@ def woe(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
target: Название целевой переменной
num_buck: Количество бакетов, если признак числовой
"""
agg = df.pipe(
bad_rate, feature, target, num_buck, group_size
).query("`target_sum` > 0")
agg = df.pipe(bad_rate, feature, target, num_buck, group_size).query(
"`target_sum` > 0"
)
return (
agg.assign(nums=agg["obj_cnt"].sum(), bad_nums=agg["target_sum"].sum())
.assign(woe=lambda x: _woe(x.bad_rate, x.bad_nums / x.nums))
Expand All @@ -339,7 +364,7 @@ def woe(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
)


def IV(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
def IV(df, feature: str, target: str, num_buck: int = 10, group_size: str = None):
"""
Считает Information Value для признака в модели бинарной классификации.
Expand All @@ -363,7 +388,9 @@ def IV(df, feature: str, target: str, num_buck: int=10, group_size: str=None):
)


def iv_report(df, features: str, target: str, num_buck: int=10, group_size: str=None):
def iv_report(
df, features: str, target: str, num_buck: int = 10, group_size: str = None
):
"""
Считает IV для указанных признаков в модели классификации.
Возвращает в порядке убывания кортежи из трех элементов:(признак, IV, интерпретация)
Expand Down Expand Up @@ -648,9 +675,7 @@ def woe_stab(df, feature, target, date, num_buck=3, date_freq="Q"):
.reset_index()
)

data = hv.Dataset(
agg, kdims=["bucket", date], vdims=["woe", "woe_b", "woe_u"]
)
data = hv.Dataset(agg, kdims=["bucket", date], vdims=["woe", "woe_b", "woe_u"])

confident_intervals = data.to.spread(
kdims=[date], vdims=["woe", "woe_b", "woe_u"], group="Confident Intervals"
Expand Down Expand Up @@ -717,18 +742,25 @@ def plot_gain_chart(target, predict, num_buck=10):
)

bars_gain = hv.Bars(
data, kdims=["bucket"], vdims=["bad_rate"], label="observed"
).opts(plot={"xrotation": 90, "show_legend": True}, style={"color": "yellow"})
data,
kdims=["bucket"],
vdims=["bad_rate"],
label="observed",
group="Observed Bad Rate",
)

curve_gain = hv.Curve(
data, kdims=["bucket"], vdims=["predict"], label="predicted"
).opts(plot={"xrotation": 90, "show_legend": True}, style={"color": "black"})
data,
kdims=["bucket"],
vdims=["predict"],
label="predicted",
group="Predicted Bad Rate",
)

return (
hv.Overlay([bars_gain, curve_gain])
hv.Overlay([bars_gain, curve_gain], group="Gain Chart")
.redim.label(**{"target": "Bad Rate"})
.relabel(f"HL_score = {H}")
.opts(plot={"legend_position": "top_left"})
)


Expand Down Expand Up @@ -756,7 +788,6 @@ def feature_importance(names, values, verbose=False, thr=0.05):
return val_dict


@_set_options
def plot_confusion_matrix(
cm, classes, normalize=False, title="Confusion matrix", cmap=plt.cm.Blues
) -> None:
Expand Down Expand Up @@ -785,7 +816,7 @@ def plot_confusion_matrix(
cm = cm.astype("float") / cm.sum(axis=1)[:, np.newaxis]
print("Normalized confusion matrix")
else:
print("Confusion matrix, without normalization")
print("Confusion matrix w/o normalization")

print(cm)

Expand All @@ -801,8 +832,8 @@ def plot_confusion_matrix(
)

plt.tight_layout()
plt.ylabel("Истинный класс")
plt.xlabel("Предсказанный класс")
plt.ylabel("Ground Truth class")
plt.xlabel("Predicted class")
plt.show()


Expand Down
2 changes: 1 addition & 1 deletion setup.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@

setup(
name="podlozhnyy_module",
version="2.5.1",
version="2.5.2",
description="One place for the most useful methods for work",
long_description=readme,
long_description_content_type="text/markdown",
Expand Down

0 comments on commit b6c5230

Please sign in to comment.