一维上的
模型构建
from keras import models from keras import layers from keras import regularizers model = models.Sequential() model.add(layers.convolutional.Conv1D(6, 5,strides=1, padding='valid', input_shape=(200, 1), activation="relu", name="convolution_1d_layer1")) model.add(layers.MaxPool1D(pool_size=2, strides=2, padding="valid", name="max_pooling_layer1")) model.add(layers.convolutional.Conv1D(12, 5,strides=1, padding='valid', activation="relu", name="convolution_1d_layer2")) model.add(layers.MaxPool1D(pool_size=2, strides=2, padding="valid", name="max_pooling_layer2")) model.add(layers.convolutional.Conv1D(24, 4,strides=1, padding='valid', activation="relu", name="convolution_1d_layer3")) model.add(layers.MaxPool1D(pool_size=2, strides=2, padding="valid", name="max_pooling_layer3")) model.add(layers.convolutional.Conv1D(48, 5,strides=1, padding='valid',activation="relu", name="convolution_1d_layer4")) model.add(layers.MaxPool1D(pool_size=2, strides=2, padding="valid", name="max_pooling_layer4")) model.add(layers.convolutional.Conv1D(96, 4,strides=1, padding='valid', activation="relu", name="convolution_1d_layer5")) model.add(layers.MaxPool1D(pool_size=1, strides=1, padding="valid", name="max_pooling_layer5")) model.add(layers.Flatten()) model.add(layers.Dense(256, kernel_regularizer=regularizers.l2(0.001), activation='relu', name="fc1")) model.add(layers.Dropout(0.5)) model.add(layers.Dense(16,activation='softmax'))- 五层卷基层池化层，一层全连层dropout层softmax层. ### 数据维度变化预览
#model.summary() _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= convolution_1d_layer1 (Conv1 (None, 196, 6) 36 _________________________________________________________________ max_pooling_layer1 (MaxPooli (None, 98, 6) 0 _________________________________________________________________ convolution_1d_layer2 (Conv1 (None, 94, 12) 372 _________________________________________________________________ max_pooling_layer2 (MaxPooli (None, 47, 12) 0 _________________________________________________________________ convolution_1d_layer3 (Conv1 (None, 44, 24) 1176 _________________________________________________________________ max_pooling_layer3 (MaxPooli (None, 22, 24) 0 _________________________________________________________________ convolution_1d_layer4 (Conv1 (None, 18, 48) 5808 _________________________________________________________________ max_pooling_layer4 (MaxPooli (None, 9, 48) 0 _________________________________________________________________ convolution_1d_layer5 (Conv1 (None, 6, 96) 18528 _________________________________________________________________ max_pooling_layer5 (MaxPooli (None, 6, 96) 0 _________________________________________________________________ flatten_1 (Flatten) (None, 576) 0 _________________________________________________________________ fc1 (Dense) (None, 256) 147712 _________________________________________________________________ dropout_1 (Dropout) (None, 256) 0 _________________________________________________________________ dense_1 (Dense) (None, 16) 4112 ================================================================= Total params: 177,744 Trainable params: 177,744 Non-trainable params: 0 _________________________________________________________________
制作数据集和标签集（训练，验证）
import numpy as np
import scipy.io as sio
def make_one_hot(data, num_label):
return (np.arange(num_label) == data).astype(np.integer)
def MaxMinNormalization(matrix):
Min = np.min(matrix)
Max = np.max(matrix)
mat_norm = (matrix - Min) / (Max - Min)
return mat_norm
def get_files():
dataset_path = 'data_sample_new.mat'
data_mat = sio.matlab.loadmat(dataset_path)
data_sample = np.array(data_mat['data']) # read data from dict
x_train = np.array(data_mat['x_train']) # read x_train data from dict
x_test = np.array(data_mat['x_test']) # read x_test data from dict
y_train = np.array(data_mat['y_train']) # read y_train data from dict
y_test = np.array(data_mat['y_test']) # read y_test data from dict
# 归一化
x_train = MaxMinNormalization(x_train)
x_test = MaxMinNormalization(x_test)
# onehot
y_train_onehot = make_one_hot(y_train, 16)
y_test_onehot = make_one_hot(y_test, 16)
print('x_train.shape',x_train.shape)
print(x_train[0])
print('y_train_onehot.shape',y_train_onehot.shape)
print(y_train_onehot[0])
return x_train, y_train_onehot, x_test, y_test_onehot
x_train, y_train_onehot, x_test, y_test_onehot = get_files()
- 数据集 x_train.shape (9224, 200)
- 标签集 y_train_onehot.shape (9224, 16)
### 训练与验证
from keras import optimizers
#模型编译
model.compile(optimizer='rmsprop',
loss='categorical_crossentropy',
metrics=['accuracy'])
x_train = np.reshape(x_train ,[-1,200,1])
x_test = np.reshape(x_test ,[-1,200,1])
history = model.fit(x_train,
y_train_onehot,
epochs=200,
batch_size=128,
validation_data=(x_test, y_test_onehot))
- 采用RMSprop优化器。因为SGD损失优化无效，不知道为啥······
### 评估模型
import matplotlib.pyplot as plt
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(acc)+1)
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validaton acc')
plt.title('Training and Validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and Validation loss')
plt.legend()
plt.show()
- 发现波动过大，采用滑动均值模型得到光滑曲线继续观测
def smooth_curve(points, factor=0.9):
smoothed_points = []
for point in points:
if smoothed_points:
previous = smoothed_points[-1]
smoothed_points.append(previous * factor + point * (1 - factor))
else:
smoothed_points.append(point)
return smoothed_points
smooth_loss = smooth_curve(loss)
smooth_val_loss = smooth_curve(val_loss)
plt.plot(range(1, len(smooth_loss) + 1), smooth_loss, 'bo',label='Training_loss')
plt.plot(range(1, len(smooth_val_loss) + 1), smooth_val_loss, 'b',label='Validation_loss')
plt.xlabel('Epochs')
plt.ylabel('Categorical_Loss')
plt.title('Training and Validation Loss')
plt.legend()
plt.figure()
smooth_acc = smooth_curve(acc)
smooth_val_acc = smooth_curve(val_acc)
plt.plot(range(1, len(smooth_acc) + 1), smooth_acc, 'bo',label='Training_acc')
plt.plot(range(1, len(smooth_val_acc) + 1), smooth_val_acc, 'b',label='Validation_acc')
plt.xlabel('Epochs')
plt.ylabel('Categorical_Accuracy')
plt.title('Training and Validation Accuracy')
plt.legend()
plt.show()
优化
- 在75轮左右过拟合，更改训练轮次
- 更改优化器，其他优化器试试
- 更改正则，L1，L2，或者一起用
- 重回训练与绘图观测
注：数据制作见博客mat处理系列