# 可読性とくに考えてません。
# とりあえず一通り作成したものを投下しておきます。

# ライブラリのインポート
#適当にパクってきてるところあり
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import lightgbm as lgb
from sklearn.model_selection import train_test_split
from sklearn.model_selection import KFold
from sklearn.metrics import accuracy_score as acc
from sklearn.preprocessing import MultiLabelBinarizer

import warnings
warnings.filterwarnings('ignore')

# データの読込
train = pd.read_csv("./data/train_data.csv")
test = pd.read_csv('./data/test_data.csv')

import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'

change_weapon = {
    "heroblaster_replica":  "hotblaster",
    "herobrush_replica": "hokusai", 
    "herocharger_replica":  "splatcharger",
    "heromaneuver_replica":"maneuver",
    "heroroller_replica":"splatroller",
    "heroshelter_replica":"parashelter",
    "heroshooter_replica":"sshooter",
    "heroslosher_replica":"bucketslosher",
    "herospinner_replica":"splatspinner",
    "octoshooter_replica":"sshooter"
                }

train = train.replace(change_weapon)
test = test.replace(change_weapon)

mlb = MultiLabelBinarizer()
mlb.fit([set(train['A4-weapon'].fillna("none").unique())])

train.head(3)

print(len(train), len(test))

train.y.sum()

def make_feature(df_train, df_test):
    train_num = len(df_train)
    df = pd.concat([df_train, df_test])
    
    
    #print(len(df))
    cat_cols = ["lobby-mode", "mode", "stage"]
    for c in cat_cols:
        vv, obj = pd.factorize(df[c])
        df[c] = vv
    #print(len(df))
    
    A1 = ['A1-weapon', 'A2-weapon', 'A3-weapon', 'A4-weapon']
    B1 = ['B1-weapon', 'B2-weapon', 'B3-weapon', 'B4-weapon']

    t = mlb.transform(df[A1].fillna("none")[A1].values)
    t2 = mlb.transform(df[B1].fillna("none")[B1].values)
    for i in range(t.shape[1]):
        df["A-" + mlb.classes_[i]] = t[:,i]
        df["B-" + mlb.classes_[i]] = t2[:, i]
    s = ["A", "B"]
    p = ["1", "2", "3", "4"]
    for i in s:
        for j in p:
            df[i+j+"-level"] = df[i+j+"-level"]//10
            df[i+j+"-level"] = df[i+j+"-level"].clip(0,30)

        
    return df[:train_num], df[train_num:]

tr, te = make_feature(train, test)

feature_cols = [
     'lobby-mode', 'mode', 'stage',
]
feature_cols.extend(["A-" + i for i in mlb.classes_])
feature_cols.extend(["B-" + i for i in mlb.classes_])
print(tr.shape)

#def cv_score(clf,X,y,n_splits=5,kf=StratifiedKFold(n_splits=5))->float:
    
kf=KFold(n_splits=5, random_state = 0)
score = 0
counter = 1
for train_index, valid_index in kf.split(tr, tr.y):
        train_X,valid_X = tr.loc[train_index,:].copy(),tr.loc[valid_index,:].copy()
        train_y,valid_y = tr.y.iloc[train_index],tr.y.iloc[valid_index]
        
        train_X, valid_X = make_feature(train_X, valid_X)
        
        
        clf = lgb.LGBMClassifier()
        clf.fit(train_X[feature_cols],train_y)
        preds = clf.predict(valid_X[feature_cols])
        acc_score = acc(valid_y,preds)
        print(f"fold{counter} score is :{acc_score}")
        score += acc_score
        counter += 1
print("average : ",round(score/5,5))

#提出用　全データ
tr, te = make_feature(train, test)
clf = lgb.LGBMClassifier().fit(tr[feature_cols].fillna(0),tr["y"].fillna(0))
pred_test = clf.predict(te[feature_cols])
pd.DataFrame({"id": range(len(pred_test)), "y": pred_test }).to_csv("submission.csv", index=False)

pred_test[pred_test == 1].shape

pred_test[pred_test == 0].shape

# importanceを表示する
#importance = pd.DataFrame(clf.feature_importances_,  index=feature_cols, importance = "gain")
importance = pd.DataFrame(clf.booster_.feature_importance(importance_type='gain'), index=feature_cols, columns = ["f"])

importance.sort_values("f", ascending = False).head(10)

mlb.classes_

s = ["A", "B"]
p = ["1", "2", "3", "4"]
d = []
for i in s:
    for j in p:
        d .extend(list(train[i+j+"-level"].values))

plt.figure(figsize = (30,5))
c = pd.DataFrame(d, columns = ["level"])
c.level = c.level//10
c.level.clip(0,30).value_counts().plot(kind = "bar")

c.level.max()

train