【Animal Crossing】Kaggleのあつまれ どうぶつの森データセットで遊んでみた

#データのインポート
import pandas as pd
import numpy as np
 
import matplotlib.pyplot as plt
import seaborn as sns
sns.set()
 
housewares = pd.read_csv('dataset/housewares.csv') #家具データ
print('The size of the test data:' + str(housewares.shape))


#display max_columnsを利用すると列数が省略されなくなる
pd.get_option("display.max_columns")
pd.set_option('display.max_columns', 33)

housewares.head(10)

#出力結果
The size of the test data:(3275, 32)

#NFSのデータを削除
housewares_nfsdrop=housewares[~housewares['Buy'].str.contains('NFS', na = False)]

#目的変数の型をintに変換
housewares_nfsdrop['Buy']=housewares_nfsdrop['Buy'].astype(int)

%matplotlib notebook
sns.distplot(housewares_nfsdrop['Buy'],bins=100,label='Buy')

#売値の高い安いの分類
housewares_nfsdrop['Buy-bin'] = 'cheap'
housewares_nfsdrop.loc[(housewares_nfsdrop['Buy']<=10000),'Buy-bin'] = 'expensive'
housewares_nfsdrop

#欠損値の確認
housewares_nfsdrop.isnull().sum()

#出力結果
Name                    0
Variation               0
Body Title           1870
Pattern               888
Pattern Title         888
DIY                     0
Body Customize          0
Pattern Customize       0
Kit Cost              867
Buy                     0
Sell                    0
Color 1                 0
Color 2                 0
Size                    0
Miles Price          1897
Source                  0
Source Notes           35
Version                 0
HHA Concept 1           0
HHA Concept 2           0
HHA Series              0
HHA Set                 0
Interact                0
Tag                     0
Outdoor                 0
Speaker Type            0
Lighting Type           0
Catalog                 0
Filename                0
Variant ID             42
Internal ID             0
Unique Entry ID         0
Buy-bin                 0
Name variation          0
dtype: int64

#Pattern ,Pattern Title はNodataで補間、KitCostは0で補間する。正しいかは不明。
housewares_nfsdrop['Pattern']=housewares_nfsdrop['Pattern'].fillna('NoData')
housewares_nfsdrop['Pattern Title']=housewares_nfsdrop['Pattern Title'].fillna('NoData')
housewares_nfsdrop['Kit Cost']=housewares_nfsdrop['Kit Cost'].fillna(0)

# X_trainとy_train作成
target_col ='Buy-bin'
feture_col = ['Pattern Title','DIY','Kit Cost','Sell','Size','Source','HHA Concept 1','HHA Concept 2','HHA Series','HHA Set','Interact','Tag','Outdoor','Speaker Type','Lighting Type','Catalog','Filename']

# 学習に必要な特徴量のみを保持
X_train = housewares_nfsdrop[feture_col]
y_train = housewares_nfsdrop[target_col]

#Label Encordingで離散化
from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()

for i in range(X_train .shape[1]):
　　#object変数のみ離散化
    if X_train .iloc[:,i].dtypes == object:
        
        le.fit(list(X_train .iloc[:,i].values) + list(X_train .iloc[:,i].values))
        X_train .iloc[:,i] = le.transform(list(X_train .iloc[:,i].values))

#ランダムフォレストでImportanceの計算
from sklearn.ensemble import RandomForestClassifier
 
rfc = RandomForestClassifier(random_state=0)
rfc.fit(X_train, y_train)
 
plt.figure(figsize=(20,10))
plt.barh(
    X_train.columns[np.argsort(rfc.feature_importances_)],
    rfc.feature_importances_[np.argsort(rfc.feature_importances_)],
    label='RandomForestClassifier'
)
plt.title('RandomForestClassifier feature importance')

#特徴量とBuyの関係
featurelist = ['Sell','Kit Cost','HHA Concept 1','Pattern Title','Size','HHA Concept 2']
length = len(featurelist)

df = pd.DataFrame()
%matplotlib inline

#グラフ描画
for i in range(length):

    #int,floatの時の処理
    if featurelist[i] == 'Sell':
        print('OK')
        fig = plt.figure()
        sns.regplot(y=X_train2['Buy'], x=X_train2['Sell'])
        
    elif featurelist[i] == 'Kit Cost':
        fig = plt.figure()
        sns.regplot(y=X_train2['Buy'], x=X_train2['Kit Cost'])

    #object変数の時の処理
    else:
        df = X_train2.groupby(featurelist[i]).mean()
        df.plot.bar(y=['Buy'], alpha=0.6)
        plt.xlabel(featurelist[i])
        plt.ylabel('Buy')

sns.distplot(housewares['Sell'],bins=100,label='Sell')

#対数変換
housewares['Sell_log'] = np.log(housewares['Sell'])
sns.distplot(housewares['Sell_log'],bins=100,label='Sell')

# X_train3とy_train3作成
target_col ='Sell_log'
feture_col = ['Pattern Title','DIY','Kit Cost','Size','Source','HHA Concept 1','HHA Concept 2','HHA Series','HHA Set','Interact','Tag','Outdoor','Speaker Type','Lighting Type','Catalog']
# 学習に必要な特徴量のみを保持
X_train3 = housewares[feture_col]
y_train3 = housewares[target_col]

#欠損値の補間
X_train3['Pattern Title']=X_train3['Pattern Title'].fillna('NoData')
X_train3['Kit Cost']=X_train3['Kit Cost'].fillna(0)

# カテゴリ変数の特徴量についてlabel encoding
from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()

for i in range(housewares_nfsdrop.shape[1]):
    if X_train3.iloc[:,i].dtypes == object:
        
        le.fit(list(X_train3.iloc[:,i].values) + list(X_train3.iloc[:,i].values))
        X_train3.iloc[:,i] = le.transform(list(X_train3.iloc[:,i].values))

#影響度が高かった特徴量のリスト
featurelist = ['Tag','HHA Concept 1','Kit Cost','Size','Pattern Title','HHA Concept 2']
length = len(featurelist)

df = pd.DataFrame()
#fig = plt.figure()
%matplotlib inline

for i in range(length):
    
    if featurelist[i] == 'Kit Cost':
        fig = plt.figure(figsize=(12,8))
        sns.regplot(y=housewares2['Sell'], x=housewares2['Kit Cost'])
        
    else:
        df = housewares2.groupby(featurelist[i]).mean()
        fig = plt.figure(figsize=(20,12))
        df.plot.bar(y=['Sell'], alpha=0.6,figsize=(12,8))
        plt.xlabel(featurelist[i])
        plt.ylabel('Sell')