如何使用LSTM进行序列分类？

Question

我正在使用深度流处理人类活动识别问题。

每个样本都是大小为（20,3，no_of_frames）的Matlab文件，即每帧中有20行和3列，并且帧数可能因不同样本而异。

我用0填充了所有的样本，所以它们都包含相同的帧数（比如说100）。

所以现在所有样本的大小都是（20,3,100）。

也让样本总数没有400，没有班是10

我如何安排我的数据集，从而在Keras使用LSTM。您是否还会为分类目的提出一些基本的LSTM模型？

Answer 1

Here is my code to create dataset: 

    clc; 
    clear all; 

    ctr=0; 
    longest=326; % length of longest sequence 
    gyroData = zeros(431,longest*3); 
    gyroLabel=zeros(431,1); 

    for a=1:27 
     for s=1:2:7 
      for t=1:4 
       fname = strcat('a', int2str(a), '_s', int2str(s), '_t', int2str(t),'_inertial.mat'); 
       if exist(fname) 
        load(fname); 
        d_iner = d_iner(:,1:3); 
        r = size(d_iner,1); 
        d_iner = cat(1,d_iner,zeros(longest-r,3)); % do zero padding to longest sequence 
        ctr = ctr+1; 
        d_iner = d_iner'; 
        gyroData(ctr,:,:) = d_iner(:); 
        gyroLabel(ctr,1)=a-1; 
       end 
      end 
     end 
    end 

    n1 = randperm(ctr); 

    for i=1:ctr 
     if i==1 
      x1=gyroData(n1(i),:); 
      y1=gyroLabel(n1(i),1); 
     else 
      x1=cat(1,x1,gyroData(n1(i),:)); 
      y1=cat(1,y1,gyroLabel(n1(i),1)); 
     end 
    end 

    %% 
    ctr=0; 
    gyroData = zeros(430,longest*3); 
    gyroLabel=zeros(430,1); 

    for a=1:27 
     for s=2:2:8 
      for t=1:4 
       fname = strcat('a', int2str(a), '_s', int2str(s), '_t', int2str(t),'_inertial.mat'); 
       if exist(fname) 
        load(fname); 
        d_iner = d_iner(:,1:3); 
        r = size(d_iner,1); 
        d_iner = cat(1,d_iner,zeros(longest-r,3)); % do zero padding to longest sequence 
        ctr = ctr+1; 
        d_iner = d_iner'; 
        gyroData(ctr,:,:) = d_iner(:); 
        gyroLabel(ctr,1)=a-1; 
       end 
      end 
     end 
    end 

    n1 = randperm(ctr); 

    for i=1:ctr 
     if i==1 
      x2=gyroData(n1(i),:); 
      y2=gyroLabel(n1(i),1); 
     else 
      x2=cat(1,x2,gyroData(n1(i),:)); 
      y2=cat(1,y2,gyroLabel(n1(i),1)); 
     end 
    end 

    save('inertial_padded.mat', 'x1', 'y1', 'x2', 'y2'); 


**And this is my LSTM code in Keras** 

import numpy as np 
import keras 
from keras.models import Sequential 
from keras.layers import Dense 
from keras.layers import LSTM 
import scipy.io 

# fix random seed for reproducibility 
np.random.seed(7) 

mat = scipy.io.loadmat('inertial_padded.mat') 
x_train = mat['x1'] 
y_train = mat['y1'] 
x_test = mat['x2'] 
y_test = mat['y2'] 

data_dim = 3 
timesteps = 326 
num_classes = 27 

x_train = x_train.reshape(x_train.shape[0], 326,3) 
x_test = x_test.reshape(x_test.shape[0], 326, 3) 
x_train = x_train.astype('float32') 
x_test = x_test.astype('float32') 

y_train = keras.utils.to_categorical(y_train, num_classes) 
y_test = keras.utils.to_categorical(y_test, num_classes) 

print('x_train shape:', x_train.shape) 
print(x_train.shape[0], 'train samples') 
print(x_test.shape[0], 'test samples') 


model = Sequential() 
model.add(LSTM(32, input_shape=(timesteps, data_dim), activation='sigmoid')) 
#model.add(LSTM(32, activation='tanh')) # returns a sequence of vectors of dimension 32 
#model.add(LSTM(32)) # return a single vector of dimension 32 
model.add(Dense(27, activation='softmax')) 

model.compile(loss='categorical_crossentropy', 
       optimizer='rmsprop', 
       metrics=['accuracy']) 

print(model.summary()) 
model.fit(x_train, y_train, 
      batch_size=16, epochs=25, 
      validation_data=(x_test, y_test)) 
# Final evaluation of the model 
scores = model.evaluate(X_test, y_test, verbose=0) 
print("Accuracy: %.2f%%" % (scores[1]*100))