Question

我正在尝试使用CNN-LSTM时间序列分类器预测运输方向（目前为二进制）。由于数据只能追溯到2010年，因此观察结果有限，因此无法收集更多数据。原始数据由一个二进制Y和68个X变量组成，这些变量进一步扩展为包括相同变量的50个滞后时间点，这些滞后会随着时间的推移而变化，从而给出X的维数[：，50,68]。

问题是当我拟合模型时，它迅速收敛为仅预测一个类（是的，数据不平衡65/35）。当尝试仅使用1层LSTM或简单的CNN时，它完全适合，而更复杂的模型仅返回一个类。任何建议（我尝试了不同的学习率，并增加了辍学率），请参见代码：

def to_supervised():# array[obs,lag,features] every y is denoted by todays and 30 last day values
    dataset = read_csv('alldataDests.csv', header=0, infer_datetime_format=True, parse_dates=['0'], index_col=['0'])
    lag,endTrain,startTest = 50,2000,2001
    mean = dataset.iloc[0:endTrain, 1:].mean(axis=0)
    dataset.iloc[:,1:]-=mean
    std = dataset.iloc[0:endTrain,1:].std(axis=0)
    dataset.iloc[:,1:] /= std
    x = dataset.values[:,1:]
    y = dataset.values[:,0]
    #train, test = dataset.values[0:endTrain], dataset.values[startTest:]
    # restructure into windows of lagged data
    xLag = array([x[i:i+lag] for i in range(len(x)-lag+1)])
    trainY = to_categorical(y[lag-1:endTrain])
    trainX=xLag[:endTrain-(lag-1),:,:]
    testY = to_categorical(y[startTest:])
    testX = xLag[startTest-(lag-1):,:,:]
    testX = testX[:1316,:,:]
    testY = testY[:1316]
    return trainX,trainY,testX,testY

def evaluate_model(trainX, trainY, testX, testy):
    verbose, epochs, batch_size = 1, 500, 128
    n_timesteps, n_features, n_outputs = trainX.shape[1], trainX.shape[2], trainY.shape[1]
    model = Sequential()
    model.add(Conv1D(32, 2, activation='relu', input_shape=(n_timesteps,n_features), 
                     kernel_regularizer = l2(0.02),
                     bias_regularizer=l2(0.02)))
    model.add(Conv1D(32, 2, activation='relu',kernel_regularizer = l2(0.02),
                     bias_regularizer=l2(0.02)))
    model.add(Dropout(0.5))
    model.add(MaxPooling1D())
    model.add(Flatten())
    model.add(Dense(100, activation='relu'))
    model.add(Dense(n_outputs, activation='softmax'))
    model.compile(loss='categorical_crossentropy', optimizer=optimizers.SGD(lr=0.00001), metrics=['accuracy'])
    # fit network
    model.fit(trainX, trainY, epochs=epochs, batch_size=batch_size, verbose=verbose)
    # evaluate model
    _, accuracy = model.evaluate(testX, testy, batch_size=batch_size, verbose=0)
    preds = model.predict_classes(testX, batch_size=batch_size, verbose=0)
    preds_proba = model.predict_proba(testX, batch_size=batch_size, verbose=0)
    return accuracy, preds,preds_proba

# summarize scores
def summarize_results(scores):
    print(scores)
    m, s = mean(scores), std(scores)
    print('Accuracy: %.3f%% (+/-%.3f)' % (m, s))

# run an experiment
def run_experiment(repeats=1):
    # load data
    trainX, trainY, testX, testY = to_supervised()
    # repeat experiment
    scores = list()
    for r in range(repeats):
        score, preds, preds_proba = evaluate_model(trainX, trainY, testX, testY)
        score = score * 100.0
        print('>#%d: %.3f' % (r+1, score))
        scores.append(score)
    # summarize results
    summarize_results(scores)
    return preds, preds_proba



preds, preds_proba= run_experiment()```

CNN-LSTM Ts分类器只能预测一个类别

0 个答案: