【備忘録】お弁当の売り上げ予想

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import japanize_matplotlib
import seaborn as sns
sns.set(font="IPAexGothic") #日本語フォント設定

train = pd.read_csv('./train.csv')
test  = pd.read_csv('./test.csv')

train['y'].plot()

sns.barplot(x="week", y='y', data =train)

sns.distplot(train['y'])

from scipy.optimize import curve_fit

def func(x, a, b, c):
    return a * np.exp(-b * x) + c

xs = train.index.values
ys = train['y'].values

popt, pcov = curve_fit(func, xs, ys)

a1, b1, c1 = popt
y_reg = a1 * np.exp(-b1 * xs) + c1

plt.figure()
plt.plot(train['y'])
plt.plot(y_reg)

new_y = train['y'] - y_reg
new_y = pd.DataFrame({'new_y': new_y})

new_y.plot()

train['new_y'] = train['y'] - y_reg

train['day'] = train['datetime'].apply(lambda x : int(x.split("-")[2]))

import plotly.offline as offline
offline.init_notebook_mode(connected=False)

import plotly.graph_objs as go

layout = go.Layout(
    autosize=False,
    width=950,
    height=500,
    margin=go.layout.Margin(
        l=80,
        r=50,
        b=50,
        t=10,
        pad=4),
    xaxis=dict(
        title='day',
        titlefont=dict(
            size=14,
        ),
        showticklabels=True,
        tickangle=0,
        tickfont=dict(
            size=12,
        ),
    ),
    yaxis=dict(
        title='トレンドを除いた販売数',
        titlefont=dict(
            size=14,
        ),
        showticklabels=True,
        tickfont=dict(
            size=10,
        ),

    ))

data = [go.Box( x=train['day'], y=train['new_y'] )]

fig = go.Figure(data=data, layout=layout)
offline.iplot(fig, filename='example', show_link=False, 
              config={"displaylogo":False, "modeBarButtonsToRemove":["sendDataToCloud"]})

week_encoded = train.groupby('week').new_y.median()

train['week'] = train['week'].replace(week_encoded)

 tickangle=90

data = [go.Box( x=train['name'], y=train['new_new_y'] )]


fig = go.Figure(data=data, layout=layout)
offline.iplot(fig, filename='example', show_link=False, 
              config={"displaylogo":False, "modeBarButtonsToRemove":["sendDataToCloud"]})

train['curry'] = train['name'].apply(lambda x : 1 if x.find("カレー") >=0 else 0)

data = [go.Box( x=train['curry'], y=train['new_new_y'] )]


fig = go.Figure(data=data, layout=layout)
offline.iplot(fig, filename='example', show_link=False, 
              config={"displaylogo":False, "modeBarButtonsToRemove":["sendDataToCloud"]})

popular_menu = set(train[train['new_new_y']>15].name)

{'さんま辛味焼',
 'ひやしたぬきうどん・炊き込みご飯',
 'やわらかロースのサムジョン',
 'カキフライタルタル',
 'キーマカレー',
 'クリームチーズ入りメンチ',
 'サバ焼味噌掛け',
 'チキンカレー',
 'チンジャオロース',
 'チーズハンバーグ',
 'ハンバーグ',
 'ハンバーグカレーソース',
 'ハンバーグデミソース',
 'ポークカレー',
 'ポークハヤシ',
 'ポーク味噌焼き',
 'メンチカツ',
 'ロコモコ丼',
 '五目御飯',
 '厚揚げ豚生姜炒め',
 '名古屋味噌カツ',
 '回鍋肉',
 '手作りヒレカツ',
 '手作りロースカツ',
 '海老クリーミ―クノーデル',
 '牛丼',
 '牛肉筍煮',
 '豚味噌メンチカツ',
 '酢豚',
 '酢豚orカレー',
 '鶏のカッシュナッツ炒め',
 '鶏のピリ辛焼き',
 '鶏の唐揚げおろしソース',
 '鶏の唐揚げ甘酢あん',
 '鶏の親子煮',
 '鶏チリソース'}

train['popular'] = train['name'].apply(lambda x : 1 if x in popular_menu else 0)

unpopular_menu = set(train[train['new_new_y']<-15].name)
train['unpopular'] = train['name'].apply(lambda x : 1 if x in unpopular_menu else 0)

train['kcal'] = train['kcal'].fillna(train['kcal'].mean())

data = [go.Box( x=train['remarks'], y=train['new_new_y'] )]


fig = go.Figure(data=data, layout=layout)
offline.iplot(fig, filename='example', show_link=False, 
              config={"displaylogo":False, "modeBarButtonsToRemove":["sendDataToCloud"]})

train['fun'] = train['remarks'].apply(lambda x: 1 if x=="お楽しみメニュー" else 0)

train['payday'] = train['payday'].fillna(0)

weather_word = ['快晴','晴','曇','雨','雪','雷']
row_index = []
number = 1

for x in weather_word:

    row_index = train['weather'].str.contains(x, na=False)
    train.loc[row_index, ['weather']] = number

    if number < 4:
        number += 1
    else:
        number = 4

    row_index = []

weather_encoded = train.groupby('weather').new_y.median()
train['weather'] = train['weather'].replace(weather_encoded)

train = train.drop(columns = ['precipitation'])

train['month'] = train['datetime'].apply(lambda x : int(x.split("-")[1]))
temp_mean = train.groupby('month').temperature.mean()
train['month'] =  train['month'].replace(temp_mean)
train['temp'] = train['temperature'] - train['month']

sns.pairplot(train)

base = ['soldout','kcal','day','payday','temperature','temp']
weather = ['weather']
week = ['week']
name = ['fun','curry','popular','unpopular']
annotation = ['beef','pork','chicken','fish','other','vegi','japanese','chinese',
             'western','grilled','sauteed','stewed','fried','steamed']

feature_x = base + weather + week + name
# feature_x = base + weather + week + name + annotation
feature_y = ['new_y']

data_x = train[feature_x]
data_y = train[feature_y]

from sklearn.ensemble import RandomForestRegressor

rf_reg = RandomForestRegressor(n_estimators=10000)
rf_reg = rf_reg.fit(data_x, data_y)

fti = rf_reg.feature_importances_

dic_arr = {'importance':fti, 'feature':feature_x}
pd.DataFrame(dic_arr).sort_values('importance', ascending=False).reset_index(drop=True).loc[:15, :]

import lightgbm as lgb

clf = lgb.LGBMRegressor()
clf.fit(data_x[feature_x], data_y[feature_y])
print ('Score: ',clf.score(data_x[feature_x], data_y[feature_y]))#正解率の表示

Score:  0.8703440689894268

train['new_y'] = clf.predict(data_x[feature_x])

original = train['y']
pred_y = train['new_y']
final_train = y_reg + pred_y

plt.figure()
plt.plot(original, label='original', linestyle='-')
plt.plot(final_train, label='trend + pred_y', linestyle='-')
plt.legend(loc='best')

参考記事

SIGNATE お弁当の売り上げをLightGBMで予測してみた