分類: Machine Learning

import tensorflow as tf
import datetime

mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

def create_model():
  return tf.keras.models.Sequential([
    tf.keras.layers.Flatten(input_shape=(28, 28)),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Dense(10, activation='softmax')
  ])

model = create_model()
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
# 設定log的資料夾
log_dir = "logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
# 設定tensorboard的callback
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)

model.fit(x=x_train, 
          y=y_train, 
          epochs=5, 
          validation_data=(x_test, y_test), 
          callbacks=[tensorboard_callback])

tensorboard --logdir logs/fit

from sklearn.model_selection import train_test_split
# 載入資料集
....
# 分離訓練和測試數據
X_train, X_val, y_train, y_val = train_test_split(train_images_fold, train_labels_fold, test_size=0.1, random_state=42)
model = create_model()
keras_classifier.fit(X_train, y_train, validation_data=(X_val, y_val))

predictions = keras_classifier.predict(X_test)
accuracy = accuracy_score(y_test, predictions)
accuracy_scores.append(accuracy)

incorrect_indices = np.where(predictions != y_test)[0]
incorrect_images = X_test[incorrect_indices]
incorrect_labels = y_test[incorrect_indices]
incorrect_prediction = predictions[incorrect_indices]
for i in range(len(incorrect_indices)):
    incorrect.append({"image": incorrect_images[i] ,"label": incorrect_labels[i], "pred": incorrect_prediction[i], "idx": fold_index})

import numpy as np
import matplotlib.pyplot as plt
import pathlib
from sklearn.model_selection import StratifiedKFold
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import accuracy_score
from tensorflow import keras
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_validate

img_path = 'dice3'

def class_names(path):
    return np.array(sorted([item.name for item in pathlib.Path(path).glob('*') if
                            "LICENSE.txt" != item.name]))
def create_model():
    model = tf.keras.Sequential()
    model.add(tf.keras.layers.Rescaling(1. / 255))
    model.add(tf.keras.layers.Conv2D(32, kernel_size=7, activation='relu'))
    model.add(tf.keras.layers.MaxPooling2D())
    model.add(tf.keras.layers.Conv2D(64, kernel_size=5, activation='relu'))
    model.add(tf.keras.layers.MaxPooling2D())
    model.add(tf.keras.layers.Flatten())
    model.add(tf.keras.layers.Dense(64, activation='relu'))
    model.add(tf.keras.layers.Dense(len(class_names(img_path))))

    model.compile(
        optimizer='adam',
        loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
        metrics=['accuracy'])
    return model


train_ds = tf.keras.utils.image_dataset_from_directory(
  img_path,
  seed=7,
  batch_size=32)
train_images = []
train_labels = []

for images, labels in train_ds:
    train_images.append(images.numpy())
    train_labels.append(labels.numpy())

train_images = np.concatenate(train_images, axis=0)
train_labels = np.concatenate(train_labels, axis=0)
label_encoder = LabelEncoder()
train_labels_encoded = label_encoder.fit_transform(train_labels)

# 創建KerasClassifier
keras_classifier = KerasClassifier(build_fn=create_model, epochs=5, batch_size=16)

# 定義StratifiedKFold
kfold = StratifiedKFold(n_splits=5, shuffle=True, random_state=123)

fold_index = 1
# 進行交叉驗證
accuracy_scores = []
incorrect = []
for train_indices, val_indices in kfold.split(train_images, train_labels_encoded):
    print("fold_index="+str(fold_index))
    train_images_fold = train_images[train_indices]
    train_labels_fold = train_labels_encoded[train_indices]

    X_test = train_images[val_indices]
    y_test = train_labels_encoded[val_indices]
    
    # 分離訓練和測試數據
    X_train, X_val, y_train, y_val = train_test_split(train_images_fold, train_labels_fold, test_size=0.1, random_state=42)
    model = create_model()
    keras_classifier.fit(X_train, y_train, validation_data=(X_val, y_val))

    predictions = keras_classifier.predict(X_test)
    accuracy = accuracy_score(y_test, predictions)
    accuracy_scores.append(accuracy)

    incorrect_indices = np.where(predictions != y_test)[0]
    incorrect_images = X_test[incorrect_indices]
    incorrect_labels = y_test[incorrect_indices]
    incorrect_prediction = predictions[incorrect_indices]
    for i in range(len(incorrect_indices)):
        incorrect.append({"image": incorrect_images[i] ,"label": incorrect_labels[i], "pred": incorrect_prediction[i], "idx": fold_index})
    # 印出準確率
    print("Accuracy scores:", accuracy_scores)
    print("Mean accuracy:", np.mean(accuracy_scores))
    fold_index += 1

# 顯示出錯誤的答案
images_per_page = 15
num_images_per_row = 5
num_images_per_col = 3
        
num_pages = (len(incorrect) - 1) // images_per_page + 1
for page in range(num_pages):
    start_idx = page * images_per_page
    end_idx = (page + 1) * images_per_page
    page_detail = incorrect[start_idx:end_idx]
    
    fig, axes = plt.subplots(num_images_per_col, num_images_per_row, figsize=(num_images_per_col*2, num_images_per_row*2))
    i=0
    for data in page_detail:
        image = data["image"]
        label = data["label"]
        pred = data["pred"]
        idx = data["idx"]
        row = i // num_images_per_row
        col = i % num_images_per_row
        ax = axes[row, col]
        image = image.astype(np.uint8)
        ax.imshow(image)
        ax.set_title(f"{label}->{pred}({idx})")
        i = i+1
        ax.axis("off")

    plt.tight_layout()
    #plt.suptitle("fold_index:"+str(fold_index)+"-"+str(page+1), x=0, y=1, ha='left', va='top')
    plt.show()

import tensorflow as tf
 
# 建立圖片資料集
dataset = tf.keras.utils.image_dataset_from_directory(
    'image_directory',
    labels='inferred',
    class_names=None,
    label_mode='int'
)
 
# 計算資料集的大小
dataset_size = tf.data.experimental.cardinality(dataset).numpy()
 
# 計算訓練集和測試集的大小
test_size = int(0.2 * dataset_size)
train_size = dataset_size - test_size
 
# 將資料集分割為訓練集和測試集
train_dataset = dataset.take(train_size)
test_dataset = dataset.skip(train_size)

import numpy as np
import pathlib
import tensorflow as tf
from sklearn.model_selection import LeaveOneOut,KFold
from sklearn.model_selection import StratifiedKFold
from sklearn.preprocessing import LabelEncoder
from sklearn.datasets import load_iris

img_path = 'dice3'
train_ds = tf.keras.utils.image_dataset_from_directory(
  img_path,
  seed=7,
  batch_size=32)
X = []
y = []

for images, labels in train_ds:
    X.append(images.numpy())
    y.append(labels.numpy())

X = np.concatenate(X, axis=0)
y = np.concatenate(y, axis=0)
label_encoder = LabelEncoder()
y = label_encoder.fit_transform(y)

def class_names(path):
    return np.array(sorted([item.name for item in pathlib.Path(path).glob('*') if
                            "LICENSE.txt" != item.name]))
def create_model():
    model = tf.keras.Sequential()
    model.add(tf.keras.layers.Rescaling(1. / 255))
    model.add(tf.keras.layers.Conv2D(32, kernel_size=7, activation='relu'))
    model.add(tf.keras.layers.MaxPooling2D())
    model.add(tf.keras.layers.Conv2D(64, kernel_size=5, activation='relu'))
    model.add(tf.keras.layers.MaxPooling2D())
    model.add(tf.keras.layers.Flatten())
    model.add(tf.keras.layers.Dense(64, activation='relu'))
    model.add(tf.keras.layers.Dense(len(class_names(img_path))))
    model.compile(
        optimizer='adam',
        loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
        metrics=['accuracy'])
    return model

def make_dataset(X_data,y_data,n_splits):
    def gen():
        for train_index, test_index in KFold(n_splits).split(X_data):
            X_train, X_test = X_data[train_index], X_data[test_index]
            y_train, y_test = y_data[train_index], y_data[test_index]
            yield X_train,y_train,X_test,y_test

    return tf.data.Dataset.from_generator(gen, (tf.float64,tf.float64,tf.float64,tf.float64))

dataset=make_dataset(X,y,5)


for X_train,y_train,X_test,y_test in dataset:
    print(len(X_train), len(y_train), len(X_test), len(y_test))
    model = create_model()
    model.fit(X_train, y_train, epochs=10, batch_size=32, validation_data=(X_test, y_test))

def make_dataset(X_data,y_data,n_splits):
    def gen():
        kfold = StratifiedKFold(n_splits=5, shuffle=True, random_state=123)
        for train_index, test_index in kfold.split(X_data, y_data):
            X_train, X_test = X_data[train_index], X_data[test_index]
            y_train, y_test = y_data[train_index], y_data[test_index]
            yield X_train,y_train,X_test,y_test

    return tf.data.Dataset.from_generator(gen, (tf.float64,tf.float64,tf.float64,tf.float64))

    # Importing the library
    import psutil


    # Getting % usage of virtual_memory ( 3rd field)
    if psutil.virtual_memory()[2] > 80:
        print(time.strftime("%Y-%m-%d_%H-%M-%S") + ' RAM memory % used:', psutil.virtual_memory()[2])
        # Getting usage of virtual_memory in GB ( 4th field)
        print('RAM Used (GB):', psutil.virtual_memory()[3]/1000000000)
    
    cpu_usage = psutil.cpu_percent()
    if cpu_usage > 80:
        print(time.strftime("%Y-%m-%d_%H-%M-%S") + ' The CPU usage is: ', cpu_usage)

gpus = tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(gpu, True)