NBA 선수 은퇴나이 추청 해보기_07

클래스 만들기

아킬레스, 십자인대 df 만들기

class AclAndAchilles:
    def __init__(self, df):
        self.df = df
        self.yes_no_df()
        self.seasonout_df()
        self.yes_no2_df()
        self.one_or_zero_df()

    # 우선은 ACL과 아킬레스가 들어간  선수 구분
    @staticmethod
    def yes_no(x):
        words = x.split(' ')
        print(words)
        for word in words:
            if word.upper() in ['ACL', 'PCL', 'ACHILLES']:
                return True
                break   

    def yes_no_df(self):
        self.df['tf'] =  self.df['Notes2'].apply(lambda x:self.yes_no(x))

    # 시즌아웃 부상
    @staticmethod
    def seasonout(x):
        words = re.split('\(|\)', x)
        print(words)
        for word in words:
            if word in ['out for season']:
                return True
                break

    def seasonout_df(self):
        self.df['out']  = self.df['Notes2'].apply(lambda x:self.seasonout(x))

    # 횟수 카운트
    @staticmethod
    def yes_no2(x):
        words = x.split(' ')
        print(words)
        sum_sum = 0
        for word in words:
            if word.upper() in ['ACL', 'PCL', 'ACHILLES'] or word.upper() in['TORN','RUPTURE']:
                sum_sum += 1
            if sum_sum >= 2:
                return True
                break   

    def yes_no2_df(self):
        self.df['tf2'] = self.df['Notes2'].apply(lambda x:self.yes_no2(x))

    # 아킬레스와 십자인대 부상 전적 여부
    @staticmethod
    def one_or_zero(x):
        two=0
        three=0
        if x['out'] == True:
            two = 1
        if  x['tf2'] == True:
            three = 1
        print(two, three)    
        return  pd.Series([two,three])

    def one_or_zero_df(self):
        self.df[['outnum','tf2num']] = self.df[['out','tf2']].apply(self.one_or_zero,axis=1)

    def df1(self):
        return  self.df.groupby('Relinquished',as_index=False).agg({'outnum':'sum','tf2num':'sum'})

• 기존에 하나씩 썼던 함수들을 클래스로 묶어서 한 번에 실행하려고 한다.
• self를 받지 않는 함수들은 @staticmethod를 달아줘야 한다.

df = pd.read_csv('nba_injury_1998.csv')
test1 = AclAndAchilles(df)
test2 =test1.df1()
test2.head()

	Relinquished	outnum	tf2num
0	(James) Mike Scott	0	0
1	(William) Tony Parker	2	0
2	76ers	0	0
3	A.J. Bramlett	0	0
4	A.J. Guyton	0	0

다른 데이터와 합치기

class injury:
    def __init__(self, acl_achilles_df, injury_df, nba_all_df):
        self.acl_achilles_df = acl_achilles_df
        self.injury_df = injury_df
        self.nba_all_df = nba_all_df
        self.merge()
        self.age_func()

    def merge(self):
        self.df_merge1 = pd.merge(self.acl_achilles_df,self.injury_df,left_on='Relinquished',right_on='name').drop('name',axis=1)

    def age_func(self):
        # 평균을 구하고 모두 소수 2번쨰까지만 살리기
        self.nba_all_group = self.nba_all_d.groupby('player_name',as_index=False).mean()
        for i in range(self.nba_all_group.shape[0]):
            for i2 in range(len(list(self.nba_all_group.columns))):
                if i2 == 0:
                    continue
                elif i2 == 1:
                    self.nba_all_group.iloc[i,i2] = self.nba_all_group.iloc[i,i2].astype('int64')
                else:
                    self.nba_all_group.iloc[i,i2] = round(self.nba_all_group.iloc[i,i2],2)

        self.nba_all_group['age'] = self.nba_all_group['age'].astype('int64')

    def final_df(self):
        self.df_merge_final = pd.merge(df_merge1,nba_all_group,left_on='Relinquished',right_on='player_name',how='left').\
                                                            drop('age_y',axis=1).rename(columns={'age_x':'age'})

        self.df_merge_final.drop('player_name',axis=1,inplace=True)

        self.df_merge_final['Notes'] = self.df_merge_final['Notes'].astype(int)
        self.df_merge_final.head()

injury_df = pd.read_csv('nba_injury_merge_position.csv')
nba_all = pd.read_csv('all_seasons.csv').drop('Unnamed: 0',axis=1)
test3 = injury(test2, injury_df, nba_all)

df_final = test3.final_df()
df_final.head()

	Relinquished	outnum	tf2num	age	season	Notes	position	player_height	player_weight	gp	pts	reb	ast	net_rating	oreb_pct	dreb_pct	usg_pct	ts_pct	ast_pct
0	Aaron Brooks	0	0	35	10	9	G	182.88	73.03	64.50	8.88	1.55	2.76	-3.31	0.02	0.07	0.22	0.52	0.23
1	Aaron Gray	0	0	30	7	14	C	213.36	122.47	45.43	3.24	3.73	0.66	-4.73	0.13	0.23	0.15	0.53	0.09
2	Adam Morrison	1	1	29	3	8	F	203.20	92.99	53.67	6.07	1.80	1.17	-7.83	0.03	0.10	0.19	0.44	0.11
3	Adonal Foyle	1	0	35	12	18	C	208.28	118.88	61.08	3.78	4.48	0.44	-4.40	0.11	0.19	0.13	0.50	0.04
4	Al Harrington	0	1	35	16	15	F	205.74	112.49	61.31	12.20	5.13	1.52	-1.45	0.06	0.17	0.23	0.51	0.10

• 잘 작동한다.

Tensorflow 클래스 만들기

import

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

import pandas as pd
import numpy as np
import seaborn as sns

print(tf.__version__)

• 필요한 패키지들을 import 한다.
• 맨 처음에 같이 불러오도록 맨 위에 다시 붙여넣기 한다.

from sklearn.preprocessing import LabelEncoder
# object인 컬럼만 찾기
df_final.info()

class Encoder_df:
        def __init__(self, df):
            self.df = df
            self.labelencoder()
            self.label_add_colums()

        def labelencoder(self):
            self.encoder = LabelEncoder()
            self.encoder.fit(list(self.df['Relinquished']))
            self.digit_label_Relinquished = self.encoder.transform(self.df['Relinquished'])

            self.encoder.fit(list(self.df['position']))
            self.digit_label_position = self.encoder.transform(self.df['position'])

        def label_add_colums(self):
             # 새로운 컬럼으로 넣어주기
            self.df['Relinquished_digit'] = self.digit_label_Relinquished
            self.df['position_digit'] = self.digit_label_position
            self.df_new  = self.df.drop(['Relinquished','position'],axis=1)

        def tensorflow(self):
            self.train_set = self.df_new.sample(frac=.8, random_state=0)
            self.test_set = self.df_new.drop(self.train_set.index)
            return self.train_set, self.test_set

tensor = Encoder_df(df_final)

• 중간까지 했을 때 잘 작동한다.

class PrintDot(keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs):
        if epoch % 100 == 0:print('')
        print(',', end='')


class Tensorflow_df:
    def __init__(self, train, test, dense_cnt, name):
        self.train_set = train
        self.test_set = test
        self.dense_cnt = dense_cnt
        self.name = name
        self._sample_result = ''
        self.train_df()
        self.y_df()
        self.norm_df()
        self.model_learn()
        self.mse_print()

    def train_df(self):
        self.train_state = self.train_set.describe()
        self.train_state.pop('age')
        self.train_state = self.train_state.T

    def y_df(self):
        self.y_train = self.train_set.pop('age')
        self.y_test = self.test_set.pop('age')

    @staticmethod
    def norm(x, train_state):
        return (x - train_state['mean']) / train_state['std']

    def norm_df(self):
        self.norm_train_set = self.norm(self.train_set, self.train_state)
        self.norm_test_set = self.norm(self.test_set, self.train_state)

    def model_learn(self):
        self.model = keras.Sequential([
            layers.Dense(self.dense_cnt, activation=self.name, input_shape=[len(train_set.keys())]),
            layers.Dense(self.dense_cnt, activation=self.name),
            layers.Dense(1)
        ])

        optimizer = tf.keras.optimizers.RMSprop()
        self.model.compile(loss='mse', optimizer = optimizer, metrics=['mae', 'mse'])

        self.model.summary()

        self._sample_result = self.model.predict(self.norm_train_set)

        self.history = self.model.fit(self.norm_train_set, self.y_train, epochs=1000, validation_split=.2, verbose=0, callbacks=[PrintDot()])

    def mse_print(self):
        loss, mae, mse = self.model.evaluate(self.norm_test_set, self.y_test,verbose=1)
        print('평균 절대 오차 : ',mae)

    def plt_show(self):
        # 시각화
        self.y_pred = self.model.predict(self.norm_test_set).flatten()
        plt.scatter(self.y_test, self.y_pred)
        plt.xlim([0,plt.xlim()[1]])
        plt.ylim([0,plt.ylim()[1]])
        plt.scatter(self.y_test, self.y_pred)
        _ = plt.plot([-100,100],[-100,100])
        plt.show()


    def history_df(self):
        self.hist = pd.DataFrame(self.history.history)
        return self.hist

    @property
    def get_result(self):
        return self._sample_result

• tensorflow하는 부분만 클래스로 다시 구성하였다.

labeling = Encoder_df(df_final)
train_set, test_set = labeling.tensorflow()

tensor = Tensorflow_df(train_set, test_set, 50, 'relu')

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_6 (Dense)              (None, 50)                950       
_________________________________________________________________
dense_7 (Dense)              (None, 50)                2550      
_________________________________________________________________
dense_8 (Dense)              (None, 1)                 51        
=================================================================
Total params: 3,551
Trainable params: 3,551
Non-trainable params: 0
_________________________________________________________________

,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
31/31 [==============================] - 0s 128us/sample - loss: 31.5719 - mae: 3.9577 - mse: 31.5719
평균 절대 오차 :  3.957709

• 다음과 같이 잘 실행되었다.

result = tensor.get_result
print(result)

• 예측 결과를 실제 나이와 비교하는 df를 만들고 이제 리액트로 페이지를 만들어보자.