반도체 수율 예측

levn94·2023년 2월 16일
machine learning
0

UCI SECOM 데이터셋 반도체 양/불 예측

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import tensorflow as tf
from sklearn.metrics import confusion_matrix

# csv 파일 읽어오기
data = pd.read_csv('/content/drive/MyDrive/data/uci-secom.csv')


1567 * 592 공정 데이터와 양/불 라벨

Colinearity Removal

상관계수가 높은 feature들을 제거하여 과적합 방지

def remove_collinear_features(x, threshold):
    corr_matrix = x.corr()
    iters = range(len(corr_matrix.columns) - 1)
    drop_cols = []
    
    for i in iters:
        for j in range(i+1):
            item = corr_matrix.iloc[j:(j+1), (i+1):(i+2)]
            col = item.columns
            row = item.index
            # 절대값을 씌우는 이유는
            # corr 절대값이 높은거를 제거하면 되기 때문에
            val = abs(item.values)
            
            if val >= threshold:
                print(col.values[0], '|', row.values[0], '|', round(val[0][0], 2))
                drop_cols.append(col.values[0])
                
    drops = set(drop_cols)
    x = x.drop(columns=drops)
            
    return x

Preprocessing and spliting data

def data_preprocess(data, th = 0.70):
    
    # 결측값 모두 0으로 바꾸기
    data = data.replace(np.NaN, 0)
    
    # correlation이 threshold 이상인 feature 제거
    data = remove_collinear_features(data, th)
    
    # time column 제거하기
    data = data.drop(columns = ['Time'], axis = 1)
    
    # single value columns : 한가지 값으로 구성된 index row(정합성 없는 데이터)
    single_value_columns = [
        '13', '42', '52', '97', '141', '149', '178', '179', '186', '189', '190',
        '191', '192', '193', '194', '226', '229', '230', '231', '232', '233', '234', '235', '236',
        '237', '240', '241', '242', '243', '256', '257', '258', '259', '260', '261', '262', '263',
        '264', '265', '266', '276', '284', '313', '314', '315', '322', '325', '326', '327', '328',
        '329', '330', '364', '369', '370', '371', '372', '373', '374', '375', '378', '379', '380',
        '381', '394', '395', '396', '397', '398', '399', '400', '401', '402', '403', '404', '414',
        '422', '449', '450', '451', '458', '461', '462', '463', '464', '465', '466', '481', '498',
        '501', '502', '503', '504', '505', '506', '507', '508', '509', '512', '513', '514', '515',
        '528', '529', '530', '531', '532', '533', '534', '535', '536', '537', '538'
    ]
    
    data = data.drop(single_value_columns, axis = 1)
    
    return data

def split_x_y(data):
    
    x = data.iloc[:, :306]
    y = data['Pass/Fail']
    
    return x, y

def train_val_test_split(x, y, random_state = 42):
    x_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.2,
                                                   random_state = random_state)
    x_val, x_test, y_val, y_test = train_test_split(x_test, y_test, test_size = 0.5,
                                                   random_state = random_state)
    
    print('x_train shape :', x_train.shape)
    print('y_train shape :', y_train.shape)

    print('x_test shape :', x_test.shape)
    print('y_test shape :', y_test.shape)

    print('x_val shape :', x_val.shape)
    print('y_val shape :', y_val.shape)
    
    return x_train, x_val, x_test, y_train, y_val, y_test

def standardization(x_train, x_val, x_test):
    sc = StandardScaler()
    sc.fit(x_train)
    x_train = sc.transform(x_train)
    x_val = sc.transform(x_val)
    x_test = sc.transform(x_test)
    
    return x_train, x_val, x_test

def y_change(y):
    y_np = np.array(y)
    y_np = np.where(y_np == -1, 0, 1)
    return y_np

def cal_metric(x_test, model):
    y_pred = np.argmax(model.predict(x_test), axis = 1)
    cm = confusion_matrix(y_test, y_pred)
    print(cm)
    tp = cm[0,0]
    fn = cm[0,1]
    fp = cm[1,0]
    tn = cm[1,1]
    precision = tp / (tp + fp)
    recall = tp / (tp + fn)
    f1 = 2*precision*recall / (precision + recall)
    acc = (tp+tn) / (tp+fn+fp+tn)
    print('precision : ', precision)
    print('recall : ', recall)
    print('f1 : ', f1)
    print('acc : ', acc)
    
    return cm, precision, recall, f1, acc
data_pre = data_preprocess(data)
x, y = split_x_y(data_pre)
x_train, x_val, x_test, y_train, y_val, y_test = train_val_test_split(x, y)
x_train, x_val, x_test = standardization(x_train, x_val, x_test)
y_train, y_val, y_test = y_change(y_train), y_change(y_val), y_change(y_test)

Model Training

model = tf.keras.Sequential([
    tf.keras.layers.Dense(32, activation = 'elu',
                         input_shape = (194,)),
    tf.keras.layers.Dense(16, activation = 'elu'),
    tf.keras.layers.Dense(8, activation = 'elu'),
    tf.keras.layers.Dense(4, activation = 'elu'),
    tf.keras.layers.Dense(2, activation = 'softmax'),    
])
model.compile(optimizer = 'adam', loss = 'sparse_categorical_crossentropy')
history = model.fit(x_train, y_train,
                validation_data = (x_val, y_val),
                epochs = 100)
Epoch 1/100

40/40 [==============================] - 1s 7ms/step - loss: 0.6910 - val_loss: 0.5348

Epoch 2/100

40/40 [==============================] - 0s 3ms/step - loss: 0.3953 - val_loss: 0.2742

Epoch 3/100

40/40 [==============================] - 0s 3ms/step - loss: 0.1743 - val_loss: 0.1016

Epoch 4/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0606 - val_loss: 0.0394

Epoch 5/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0233 - val_loss: 0.0180

Epoch 6/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0119 - val_loss: 0.0108

Epoch 7/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0072 - val_loss: 0.0071

Epoch 8/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0048 - val_loss: 0.0050

Epoch 9/100

40/40 [==============================] - 0s 4ms/step - loss: 0.0034 - val_loss: 0.0037

Epoch 10/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0025 - val_loss: 0.0030

Epoch 11/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0020 - val_loss: 0.0023

Epoch 12/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0016 - val_loss: 0.0019

Epoch 13/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0013 - val_loss: 0.0016

Epoch 14/100

40/40 [==============================] - 0s 3ms/step - loss: 0.0011 - val_loss: 0.0014

Epoch 15/100

40/40 [==============================] - 0s 3ms/step - loss: 9.0052e-04 - val_loss: 0.0012

Epoch 16/100

40/40 [==============================] - 0s 3ms/step - loss: 7.7080e-04 - val_loss: 0.0010

Epoch 17/100

40/40 [==============================] - 0s 3ms/step - loss: 6.6596e-04 - val_loss: 9.0644e-04

Epoch 18/100

40/40 [==============================] - 0s 3ms/step - loss: 5.8166e-04 - val_loss: 8.0559e-04

Epoch 19/100

40/40 [==============================] - 0s 3ms/step - loss: 5.1117e-04 - val_loss: 7.1926e-04

Epoch 20/100

40/40 [==============================] - 0s 3ms/step - loss: 4.5259e-04 - val_loss: 6.4941e-04

Epoch 21/100

40/40 [==============================] - 0s 3ms/step - loss: 4.0332e-04 - val_loss: 5.8738e-04

Epoch 22/100

40/40 [==============================] - 0s 3ms/step - loss: 3.6201e-04 - val_loss: 5.3342e-04

Epoch 23/100

40/40 [==============================] - 0s 3ms/step - loss: 3.2702e-04 - val_loss: 4.8694e-04

Epoch 24/100

40/40 [==============================] - 0s 3ms/step - loss: 2.9607e-04 - val_loss: 4.4644e-04

Epoch 25/100

40/40 [==============================] - 0s 3ms/step - loss: 2.6936e-04 - val_loss: 4.1285e-04

Epoch 26/100

40/40 [==============================] - 0s 3ms/step - loss: 2.4583e-04 - val_loss: 3.8190e-04

Epoch 27/100

40/40 [==============================] - 0s 3ms/step - loss: 2.2549e-04 - val_loss: 3.5411e-04

Epoch 28/100

40/40 [==============================] - 0s 3ms/step - loss: 2.0717e-04 - val_loss: 3.2888e-04

Epoch 29/100

40/40 [==============================] - 0s 3ms/step - loss: 1.9080e-04 - val_loss: 3.0687e-04

Epoch 30/100

40/40 [==============================] - 0s 3ms/step - loss: 1.7645e-04 - val_loss: 2.8686e-04

Epoch 31/100

40/40 [==============================] - 0s 3ms/step - loss: 1.6355e-04 - val_loss: 2.6849e-04

Epoch 32/100

40/40 [==============================] - 0s 3ms/step - loss: 1.5222e-04 - val_loss: 2.5245e-04

Epoch 33/100

40/40 [==============================] - 0s 3ms/step - loss: 1.4176e-04 - val_loss: 2.3701e-04

Epoch 34/100

40/40 [==============================] - 0s 3ms/step - loss: 1.3223e-04 - val_loss: 2.2391e-04

Epoch 35/100

40/40 [==============================] - 0s 2ms/step - loss: 1.2371e-04 - val_loss: 2.1115e-04

Epoch 36/100

40/40 [==============================] - 0s 3ms/step - loss: 1.1584e-04 - val_loss: 1.9947e-04

Epoch 37/100

40/40 [==============================] - 0s 3ms/step - loss: 1.0868e-04 - val_loss: 1.8877e-04

Epoch 38/100

40/40 [==============================] - 0s 3ms/step - loss: 1.0212e-04 - val_loss: 1.7868e-04

Epoch 39/100

40/40 [==============================] - 0s 3ms/step - loss: 9.5822e-05 - val_loss: 1.6915e-04

Epoch 40/100

40/40 [==============================] - 0s 2ms/step - loss: 9.0310e-05 - val_loss: 1.6053e-04

Epoch 41/100

40/40 [==============================] - 0s 3ms/step - loss: 8.5121e-05 - val_loss: 1.5296e-04

Epoch 42/100

40/40 [==============================] - 0s 3ms/step - loss: 8.0457e-05 - val_loss: 1.4585e-04

Epoch 43/100

40/40 [==============================] - 0s 2ms/step - loss: 7.6151e-05 - val_loss: 1.3880e-04

Epoch 44/100

40/40 [==============================] - 0s 3ms/step - loss: 7.2066e-05 - val_loss: 1.3265e-04

Epoch 45/100

40/40 [==============================] - 0s 3ms/step - loss: 6.8286e-05 - val_loss: 1.2683e-04

Epoch 46/100

40/40 [==============================] - 0s 3ms/step - loss: 6.4747e-05 - val_loss: 1.2130e-04

Epoch 47/100

40/40 [==============================] - 0s 3ms/step - loss: 6.1486e-05 - val_loss: 1.1619e-04

Epoch 48/100

40/40 [==============================] - 0s 2ms/step - loss: 5.8486e-05 - val_loss: 1.1126e-04

Epoch 49/100

40/40 [==============================] - 0s 3ms/step - loss: 5.5665e-05 - val_loss: 1.0677e-04

Epoch 50/100

40/40 [==============================] - 0s 3ms/step - loss: 5.3016e-05 - val_loss: 1.0232e-04

Epoch 51/100

40/40 [==============================] - 0s 3ms/step - loss: 5.0544e-05 - val_loss: 9.8180e-05

Epoch 52/100

40/40 [==============================] - 0s 3ms/step - loss: 4.8187e-05 - val_loss: 9.4617e-05

Epoch 53/100

40/40 [==============================] - 0s 3ms/step - loss: 4.6015e-05 - val_loss: 9.0769e-05

Epoch 54/100

40/40 [==============================] - 0s 3ms/step - loss: 4.3958e-05 - val_loss: 8.7460e-05

Epoch 55/100

40/40 [==============================] - 0s 3ms/step - loss: 4.2015e-05 - val_loss: 8.4257e-05

Epoch 56/100

40/40 [==============================] - 0s 3ms/step - loss: 4.0169e-05 - val_loss: 8.1055e-05

Epoch 57/100

40/40 [==============================] - 0s 3ms/step - loss: 3.8458e-05 - val_loss: 7.8214e-05

Epoch 58/100

40/40 [==============================] - 0s 3ms/step - loss: 3.6821e-05 - val_loss: 7.5422e-05

Epoch 59/100

40/40 [==============================] - 0s 3ms/step - loss: 3.5285e-05 - val_loss: 7.2857e-05

Epoch 60/100

40/40 [==============================] - 0s 3ms/step - loss: 3.3845e-05 - val_loss: 7.0336e-05

Epoch 61/100

40/40 [==============================] - 0s 3ms/step - loss: 3.2463e-05 - val_loss: 6.7829e-05

Epoch 62/100

40/40 [==============================] - 0s 3ms/step - loss: 3.1151e-05 - val_loss: 6.5504e-05

Epoch 63/100

40/40 [==============================] - 0s 2ms/step - loss: 2.9924e-05 - val_loss: 6.3517e-05

Epoch 64/100

40/40 [==============================] - 0s 3ms/step - loss: 2.8765e-05 - val_loss: 6.1475e-05

Epoch 65/100

40/40 [==============================] - 0s 3ms/step - loss: 2.7640e-05 - val_loss: 5.9419e-05

Epoch 66/100

40/40 [==============================] - 0s 3ms/step - loss: 2.6579e-05 - val_loss: 5.7525e-05

Epoch 67/100

40/40 [==============================] - 0s 3ms/step - loss: 2.5584e-05 - val_loss: 5.5709e-05

Epoch 68/100

40/40 [==============================] - 0s 3ms/step - loss: 2.4613e-05 - val_loss: 5.3930e-05

Epoch 69/100

40/40 [==============================] - 0s 3ms/step - loss: 2.3707e-05 - val_loss: 5.2189e-05

Epoch 70/100

40/40 [==============================] - 0s 3ms/step - loss: 2.2833e-05 - val_loss: 5.0672e-05

Epoch 71/100

40/40 [==============================] - 0s 2ms/step - loss: 2.2004e-05 - val_loss: 4.9135e-05

Epoch 72/100

40/40 [==============================] - 0s 3ms/step - loss: 2.1209e-05 - val_loss: 4.7582e-05

Epoch 73/100

40/40 [==============================] - 0s 3ms/step - loss: 2.0439e-05 - val_loss: 4.6201e-05

Epoch 74/100

40/40 [==============================] - 0s 2ms/step - loss: 1.9710e-05 - val_loss: 4.4798e-05

Epoch 75/100

40/40 [==============================] - 0s 3ms/step - loss: 1.9017e-05 - val_loss: 4.3539e-05

Epoch 76/100

40/40 [==============================] - 0s 3ms/step - loss: 1.8355e-05 - val_loss: 4.2272e-05

Epoch 77/100

40/40 [==============================] - 0s 3ms/step - loss: 1.7726e-05 - val_loss: 4.1108e-05

Epoch 78/100

40/40 [==============================] - 0s 3ms/step - loss: 1.7102e-05 - val_loss: 3.9814e-05

Epoch 79/100

40/40 [==============================] - 0s 3ms/step - loss: 1.6518e-05 - val_loss: 3.8695e-05

Epoch 80/100

40/40 [==============================] - 0s 2ms/step - loss: 1.5940e-05 - val_loss: 3.7577e-05

Epoch 81/100

40/40 [==============================] - 0s 3ms/step - loss: 1.5397e-05 - val_loss: 3.6559e-05

Epoch 82/100

40/40 [==============================] - 0s 3ms/step - loss: 1.4873e-05 - val_loss: 3.5606e-05

Epoch 83/100

40/40 [==============================] - 0s 3ms/step - loss: 1.4383e-05 - val_loss: 3.4558e-05

Epoch 84/100

40/40 [==============================] - 0s 3ms/step - loss: 1.3916e-05 - val_loss: 3.3604e-05

Epoch 85/100

40/40 [==============================] - 0s 3ms/step - loss: 1.3460e-05 - val_loss: 3.2669e-05

Epoch 86/100

40/40 [==============================] - 0s 3ms/step - loss: 1.3013e-05 - val_loss: 3.1857e-05

Epoch 87/100

40/40 [==============================] - 0s 3ms/step - loss: 1.2596e-05 - val_loss: 3.1007e-05

Epoch 88/100

40/40 [==============================] - 0s 2ms/step - loss: 1.2200e-05 - val_loss: 3.0115e-05

Epoch 89/100

40/40 [==============================] - 0s 3ms/step - loss: 1.1807e-05 - val_loss: 2.9350e-05

Epoch 90/100

40/40 [==============================] - 0s 2ms/step - loss: 1.1440e-05 - val_loss: 2.8617e-05

Epoch 91/100

40/40 [==============================] - 0s 3ms/step - loss: 1.1083e-05 - val_loss: 2.7919e-05

Epoch 92/100

40/40 [==============================] - 0s 3ms/step - loss: 1.0741e-05 - val_loss: 2.7211e-05

Epoch 93/100

40/40 [==============================] - 0s 3ms/step - loss: 1.0411e-05 - val_loss: 2.6522e-05

Epoch 94/100

40/40 [==============================] - 0s 4ms/step - loss: 1.0091e-05 - val_loss: 2.5902e-05

Epoch 95/100

40/40 [==============================] - 0s 3ms/step - loss: 9.7864e-06 - val_loss: 2.5225e-05

Epoch 96/100

40/40 [==============================] - 0s 3ms/step - loss: 9.4818e-06 - val_loss: 2.4528e-05

Epoch 97/100

40/40 [==============================] - 0s 2ms/step - loss: 9.1961e-06 - val_loss: 2.3941e-05

Epoch 98/100

40/40 [==============================] - 0s 2ms/step - loss: 8.9172e-06 - val_loss: 2.3297e-05

Epoch 99/100

40/40 [==============================] - 0s 3ms/step - loss: 8.6489e-06 - val_loss: 2.2740e-05

Epoch 100/100

40/40 [==============================] - 0s 3ms/step - loss: 8.3913e-06 - val_loss: 2.2201e-05

Performance


from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from sklearn.metrics import confusion_matrix

print("Accuracy: ", accuracy_score(y_test, y_pred)*100)
y_pred = np.argmax(model.predict(x_test), axis =1)
print('f1 score : ', f1_score(y_test, y_pred))

cm = confusion_matrix(y_test, y_pred)
sns.heatmap(cm, annot = True)

profile
levn94
Data Science & Engineering
이전 포스트

Numpy로 NeuralNet 구현하기

다음 포스트

최대공약수

2개의 댓글

comment-user-thumbnail
이원희
2023년 11월 7일

글 잘 봤습니다!!
성능이 굉장히 좋아서 저도 사용해보려 했는데 제 생각에 오류가 있는 것 같아 물어봅니다
def split_x_y(data):

x = data.iloc[:, :306]
y = data['Pass/Fail']

return x, y

이 부분에서 x안에 Pass_Fail column이 포함되어 있어 성능이 잘나오는가 싶어 어떻게 생각하시는지 궁금하네요

Pass/Fail부분 빼고 실험해보니 TN부분은 낮아지고 FP부분이 높아져 어떻게 생각하시는지 궁금합니다

1개의 답글