def transform(self, X):
log(f"start TP transform")
for i in range(len(self.exec_cols)):
cat_col = self.exec_cols[i][0]
new_col = self.new_cols[i]
X[new_col] = downcast(X[cat_col].map(self.mean_map_dict[new_col]),
accuracy_loss=False)
def func(df):
col = df.columns[0]
num_col = df.columns[1]
df[num_col] = df[num_col].astype('float32')
means = df.groupby(col, sort=False)[num_col].mean()
return tuple(df.columns), downcast(means)
def mean_label(ss: pd.Series, y):
col = ss.name
df = pd.concat([ss, y], axis=1)
df.columns = [col, 'label']
df = pd.concat([
df[col].value_counts(), df.loc[df['label'] == 1,
col].value_counts()
],
axis=1)
df.columns = ['cnt', 'pos_cnt']
df['rate'] = df['cnt'] / df['pos_cnt'] * 1.0
return col, downcast(df['rate'], accuracy_loss=False)
def transform(self, X: pd.DataFrame):
log(f"start TP transform")
def func(df):
cats = gen_combine_cats(df, df.columns)
return tuple(df.columns), cats
res = Parallel(n_jobs=CONSTANT.JOBS,
require='sharedmem')(delayed(func)(X[[col1, col2]])
for col1, col2 in self.exec_cols)
for cols, cats in res:
new_col = self.new_cols[self.exec_cols.index(cols)]
X[new_col] = downcast(cats.map(self.cnt_map_dict[new_col]),
accuracy_loss=False)
def func(df):
cats = gen_combine_cats(df, df.columns)
cnt = cats.value_counts()
return tuple(df.columns), downcast(cnt)
def values_cnt(ss: pd.Series):
counts = ss.value_counts()
return ss.name, downcast(counts)
def time_atr(ss: pd.Series, atr):
return atr, downcast(getattr(ss.dt, atr), accuracy_loss=False)
def func(df):
cols = list(df.columns)
diff = df[cols[0]] - df[cols[1]]
return tuple(df.columns), downcast(diff, accuracy_loss=False)
def values_cnt(ss: pd.Series):
counts = ss.groupby(ss).cumcount()
return ss.name, downcast(counts)