Kaggle Masterに学ぶ機械学習実践アプローチ 写経 06

# %%
import pandas as pd 
import numpy as np

# %%
s = pd.date_range('2020-01-06', '2020-01-10', freq='10h').to_series()

# %%
s

# %%
features = {
"dayofweek": s.dt.dayofweek.values,
"dayofyear": s.dt.dayofyear.values,
"hour": s.dt.hour.values,
"is_leap_year": s.dt.is_leap_year.values,
"quarter": s.dt.quarter.values
}

# %%
features

# %%
def generate_features(df):
    df.loc[:,'year'] = df["date"].dt.year
    # df.loc[:,'weekofyear'] = df["date"].dt.weekofyear
    df.loc[:,'month'] = df["date"].dt.month
    df.loc[:,'dayofweek'] = df["date"].dt.dayofweek
    df.loc[:,'weekend'] = (df["date"].dt.weekday >= 5).astype(int)

    # create an aggregate dictionary 
    aggs = {}
    aggs['month'] = ['nunique', 'mean']
    # aggs['weekofyear'] = ['nunique', 'mean']
    aggs['num1'] = ['sum', 'max', 'min', 'mean']
    aggs['customer_id'] = ['size']
    aggs['customer_id'] = ['nunique']
    print(aggs)

    agg_df = df.groupby('customer_id').agg(aggs)
    agg_df= agg_df.reset_index()
    return agg_df

# %%
df = pd.DataFrame({'date': pd.date_range('2020-01-06', '2020-01-10', freq='10h')})
df['num1'] = np.random.randint(0, 100, size=len(df))  # ダミーデータを追加
df['customer_id'] = np.random.randint(1, 10, size=len(df))  # ダミーの顧客IDを追加

df_w_features = generate_features(df)

# %%
df_w_features

# %%
import numpy as np 

# %%
x = [1,2,3,4,5,6,7,8,9,10]

# %%
feature_dict = {}

# 平均
feature_dict['mean'] = np.mean(x)
feature_dict['max'] = np.max(x)
feature_dict['min'] = np.min(x)
feature_dict['std'] = np.std(x)
feature_dict['var'] = np.var(x)
#最小値と最大値の差
feature_dict['ptp'] = np.ptp(x)

# パーセンタイル
feature_dict['percentile_10'] = np.percentile(x, 10)
feature_dict['percentile_60'] = np.percentile(x, 60)
feature_dict['percentile_90'] = np.percentile(x, 90)

# 中央値
feature_dict['quantile_5'] = np.quantile(x, 0.05)
feature_dict['quantile_95'] = np.quantile(x, 0.95)
feature_dict['quantile_99'] = np.quantile(x, 0.99)

feature_dict

# %%
print(feature_dict)

# %%
from tsfresh.feature_extraction import feature_calculators as fc

# %%
feature_dict['abs_energy'] = fc.abs_energy(x)
feature_dict['count_above_mean'] = fc.count_above_mean(x)
feature_dict['count_below_mean'] = fc.count_below_mean(x)
feature_dict['mean_abs_change'] = fc.mean_abs_change(x)
feature_dict['mean_change'] = fc.mean_change(x)


# %%
feature_dict

# %%
df = pd.DataFrame(np.random.rand(100,2), columns=[f"f_{i}" for i in range(1,3)])

# %%
df

# %%
from sklearn import preprocessing

# %%
pf = preprocessing.PolynomialFeatures(degree=2, interaction_only=False, include_bias=False)

# %%
pf.fit(df)

# %%
poly_feats = pf.transform(df)

# %%
poly_feats

# %%
num_feats = poly_feats.shape[1]
num_feats

# %%
df_transformed = pd.DataFrame(poly_feats, columns=[f"f_{i}" for i in range(1, num_feats+1)])
df_transformed

# %%
print(0.489843*0.084497)

# %%
# 量的変数の列を10つのbinに分割
df["f_bin_10"] = pd.cut(df["f_1"], bins=10, labels=False)
df["f_bin_100"] = pd.cut(df["f_1"], bins=100, labels=False)

# %%
df

# %%
df["f_3"] = pd.DataFrame(np.random.rand(10000,1)*10000, columns=["f_3"])

# %%
df

# %%
df["f_3_log"] = np.log(df["f_3"] + 1)

# %%
df

# %%
import numpy as np
from sklearn import impute

# %%
X = np.random.randint(1,15, (10,6))

# %%
X

# %%
X = X.astype(float)

# %%
X

# %%
X.ravel()[np.random.choice(X.size, 10, replace=False)] = np.nan

# %%
X

# %%
knn_imputer = impute.KNNImputer(n_neighbors=2)
knn_imputer.fit_transform(X)

# %%