TensorFlow 中的一维 CNN 用于时间序列分类

Question

我的时间序列是一个 30000 x 500 的表，表示来自三种不同类型图形的点：线性、二次和三次正弦曲线。因此，线性图有 10000 行，二次图有 10000 行，三次图有 10000 行。我从每张图中采样了 500 个点。这是一张图片来说明我的观点：

我已经使用 TensorFlow 构建了 98% 准确度的 2D CNN，但现在我想使用 TensorFlow 构建 1D CNN。我只需将每Conv2D一层替换为吗Conv1D？如果是这样，我的filters和kernel_size会是什么？我什至不知道如何导入我的一维熊猫数据框。我的 2D CNN 具有以下架构：

model = tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.Rescaling(1./255),
  tf.keras.layers.Conv1D( 32, 3, activation='relu', input_shape=input_shape[2:])(x), #32 FILTERS and square stride of size 3
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Conv2D(32, 3, activation='relu'),
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Conv2D(32, 3, activation='relu'),
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Flatten(),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dense(num_classes)
])

如果有人可以提供帮助，那就太好了。谢谢。下面是 MWE，我的 2D CNN 在这里。

num_classes = 3
model = tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.Rescaling(1./255),
  tf.keras.layers.Conv2D(32, 3, activation='relu'), #32 FILTERS and square stride of size 3
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Conv2D(32, 3, activation='relu'),
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Conv2D(32, 3, activation='relu'),
  tf.keras.layers.MaxPooling2D(),
  tf.keras.layers.Flatten(),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dense(num_classes)
])


epochs = 5

initial_learning_rate = 1
decay = initial_learning_rate / epochs

def lr_time_based_decay(epoch, lr):
    return lr * 1 / (1 + decay * epoch)

history = model.fit(
  train_ds,
  validation_data=val_ds,
  epochs= epochs,
  callbacks= [tensorboard_callback, tf.keras.callbacks.LearningRateScheduler(lr_time_based_decay, verbose=1)]
)

Answer 1

Conv1D 等效代码。Conv1D 层需要 3D 输入并输出 3D 形状。Maxpooling2D 需要 4D 输入。您需要使用 maxpooling1D 层。

示例代码

import tensorflow as tf
input_shape = (4, 7, 10, 128)
num_classes = 3

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv1D(filters= 32, kernel_size=3, activation='relu',padding='same',input_shape= input_shape[2:]))
model.add(tf.keras.layers.MaxPooling1D())
model.add(tf.keras.layers.Conv1D(filters=32, kernel_size=3,padding='same',activation='relu'))
model.add(tf.keras.layers.MaxPooling1D())
model.add(tf.keras.layers.Conv1D(filters=32, kernel_size=3,padding='same',activation='relu'))
model.add(tf.keras.layers.MaxPooling1D())
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dense(num_classes, activation='softmax'))

model.summary()

输出

Model: "sequential_13"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv1d_35 (Conv1D)          (None, 10, 32)            12320     
                                                                 
 max_pooling1d_20 (MaxPoolin  (None, 5, 32)            0         
 g1D)                                                            
                                                                 
 conv1d_36 (Conv1D)          (None, 5, 32)             3104      
                                                                 
 max_pooling1d_21 (MaxPoolin  (None, 2, 32)            0         
 g1D)                                                            
                                                                 
 conv1d_37 (Conv1D)          (None, 2, 32)             3104      
                                                                 
 max_pooling1d_22 (MaxPoolin  (None, 1, 32)            0         
 g1D)                                                            
                                                                 
 flatten_8 (Flatten)         (None, 32)                0         
                                                                 
 dense_15 (Dense)            (None, 128)               4224      
                                                                 
 dense_16 (Dense)            (None, 3)                 387       
                                                                 
=================================================================
Total params: 23,139
Trainable params: 23,139
Non-trainable params: 0