인공지능과 설계: 해석 예측에서 설계 최적화까지
Prof. Seungchul Lee
http://iailab.kaist.ac.kr/
Industrial AI Lab at KAIST
http://iailab.kaist.ac.kr/
Industrial AI Lab at KAIST
Practice Aims & Objectives¶
- Implement convolution neural networks (CNN) for classification tasks
- Perform pre-processing for image data
In [ ]:
import numpy as np
import pandas as pd
import tensorflow as tf
import matplotlib.pyplot as plt
import cv2
%matplotlib inline
import sklearn
from sklearn.model_selection import train_test_split
# import warnings
# import os
# from os.path import join
1.2 Images¶
1.3 Convolution on Image (= Convolution in 2D)¶
Filter (or Kernel)
- Modify or enhance an image by filtering
- Filter images to emphasize certain features or remove other features
Filtering includes smoothing, sharpening and edge enhancement
Discrete convolution can be viewed as element-wise multiplication by a matrix
How to find the right Kernels
We learn many different kernels that make specific effect on images
Let’s apply an opposite approach
We are not designing the kernel, but are learning the kernel from data
Can learn feature extractor from data using a deep learning framework
2.2 Load Data¶
In [ ]:
from google.colab import drive
drive.mount('/content/drive')
In [ ]:
df = pd.read_pickle('/content/drive/MyDrive/tutorials/산학협동강좌/data/wafers.pkl')
# USB로 파일을 다운 받으신 경우
# df = pd.read_pickle('wafers.pkl')
In [ ]:
df.head()
Out[ ]:
In [ ]:
print('Number of Center Failures: {}'.format(len(np.where(df.encoded_labels == 0)[0])))
print('Number of Edge-Local Failures: {}'.format(len(np.where(df.encoded_labels == 1)[0])))
print('Number of Edge-Ring Failures: {}'.format(len(np.where(df.encoded_labels == 2)[0])))
print('Number of Local Failures: {}'.format(len(np.where(df.encoded_labels == 3)[0])))
2.3 Image Pre-processing¶
Different Image Size → Image Resizing
In [ ]:
random_indices = np.random.choice(range(1200), 20, replace=False)
plt.figure(figsize = (10, 10))
for i in range(20):
ax = plt.subplot(4, 5, i + 1)
ax.axis('off')
ax.imshow(df['waferMap'][random_indices[i]])
ax.set_title('Shape: {}'.format(df['waferMap'][random_indices[i]].shape))
Resizing for Same Size
In [ ]:
resized_images = []
for i in range(len(df)):
img = df.iloc[i]['waferMap']
resized_img = cv2.resize(img, (32, 32))
resized_images.append(resized_img)
df['waferMap_resized'] = resized_images
In [ ]:
plt.figure(figsize = (10, 10))
for i in range(20):
ax = plt.subplot(4, 5, i + 1)
ax.axis('off')
ax.imshow(df['waferMap_resized'][random_indices[i]])
ax.set_title('Shape: {}'.format(df['waferMap_resized'][random_indices[i]].shape))
Data Split: Train/Test
In [ ]:
all_x = df['waferMap_resized'].values
all_y = df['encoded_labels'].values
train_x, test_x, train_y, test_y = train_test_split(all_x, all_y, test_size = 0.2, random_state = 42)
In [ ]:
print(train_y.shape)
print(test_y.shape)
In [ ]:
train_x = np.stack(train_x).reshape(-1, 32, 32, 1)
train_y = np.stack(train_y)
test_x = np.stack(test_x).reshape(-1, 32, 32, 1)
test_y = np.stack(test_y)
In [ ]:
print('Train x : {}, y : {}'.format(train_x.shape, train_y.shape))
print('Test x: {}, y : {}'.format(test_x.shape, test_y.shape))
2.4 Train and Results¶
Build CNN Model
In [ ]:
model = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(32, (3, 3),
activation = 'relu',
padding = 'SAME',
input_shape = (32, 32, 1)),
tf.keras.layers.MaxPool2D((2, 2)),
tf.keras.layers.Conv2D(64, (3, 3),
activation = 'relu',
padding = 'SAME',
input_shape = (14, 14, 32)),
tf.keras.layers.MaxPool2D((2, 2)),
tf.keras.layers.Flatten(input_shape = (7, 7)),
tf.keras.layers.Dense(32, activation = 'relu'),
tf.keras.layers.Dense(4, activation = 'softmax')
])
model.summary()
Training
In [ ]:
model.compile(optimizer = 'adam',
loss = 'sparse_categorical_crossentropy',
metrics = ['accuracy'])
In [ ]:
history = model.fit(train_x, train_y,
epochs = 10,
batch_size = 32)
In [ ]:
history.history.keys()
Out[ ]:
Learning Curve
In [ ]:
# accuracy plot
plt.plot(history.history['accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.show()
# loss plot
plt.plot(history.history['loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.show()
Test Result
In [ ]:
test_loss, test_acc = model.evaluate(test_x, test_y, verbose = 2)
print('loss = {}, Accuracy = {} %'.format(test_loss, test_acc * 100))
In [ ]:
pred_y = model.predict_on_batch(test_x[:4])
for i in range(4):
plt.figure(figsize = (8, 4))
plt.subplot(1, 2, 1)
plt.imshow(test_x[i].reshape(32, 32))
plt.axis('off')
plt.subplot(1, 2, 2)
plt.stem(pred_y[i])
plt.xticks(np.arange(4), ['Center', 'Edge-Loc', 'Edge-Ring', 'Loc']) # Set text labels.
plt.show()
print('Prediction : {}'.format(np.argmax(pred_y[i])))
print('Probability : {}'.format(pred_y[i]))
Confusion Matrix
In [ ]:
import seaborn as sns
from sklearn.metrics import confusion_matrix
#Predict
pred_y = model.predict(test_x)
pred_y = np.argmax(pred_y, axis = 1)
result = confusion_matrix(test_y, pred_y, normalize = 'pred')
print(result)
In [ ]:
plt.figure(figsize = (6, 5))
heatmap = sns.heatmap(result, annot = True)
heatmap.yaxis.set_ticklabels(['Center', 'Edge-Loc', 'Edge-Ring', 'Loc'], fontsize = 10)
heatmap.xaxis.set_ticklabels(['Center', 'Edge-Loc', 'Edge-Ring', 'Loc'], fontsize = 10)
plt.ylabel('True label', fontsize = 13)
plt.xlabel('Predicted label', fontsize = 13)
plt.show()
