Tflearn 网络架构未收敛

Question

在获得了很多帮助后，我最后一次来这里解决我找不到解决方案的最后一个问题。

在我之前的question 之后，一位用户指出我的时间序列预测结果不佳可能是因为我的架构没有收敛。

在查看并尝试了一些我在其他问题上发现的修复（设置权重、降低学习率、更改优化器/激活）之后，我似乎无法获得更好的结果，总是获得 0（或 0.0003）的准确度，这还不够好）。

我的代码：

import numpy
import numpy as np
import tflearn
from pandas import DataFrame
from pandas import Series
from pandas import concat
from pandas import read_csv
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from math import sqrt

import datetime

# Preprocessing function
from tflearn import Accuracy, Momentum


def preprocess(data):
    return np.array(data, dtype=np.int32)

def parser(x):
    return datetime.datetime.strptime(x, '%Y-%m-%d %H:%M:%S')


# frame a sequence as a supervised learning problem
def timeseries_to_supervised(data, lag=1):
    df = DataFrame(data)
    columns = [df.shift(i) for i in range(1, lag + 1)]
    columns.append(df)
    df = concat(columns, axis=1)
    df.fillna(0, inplace=True)
    return df


def difference(dataset, interval=1):
    diff = list()
    for i in range(interval, len(dataset)):
        value = dataset[i] - dataset[i - interval]
        diff.append(value)
    return Series(diff)


# invert differenced value
def inverse_difference(history, yhat, interval=1):
    return yhat + history[-interval]


# scale train and test data to [-1, 1]
def scale(train, test):
    # fit scaler
    scaler = MinMaxScaler(feature_range=(-1, 1))
    scaler = scaler.fit(train)
    # transform train
    train = train.reshape(train.shape[0], train.shape[1])
    train_scaled = scaler.transform(train)
    # transform test
    test = test.reshape(test.shape[0], test.shape[1])
    test_scaled = scaler.transform(test)
    return scaler, train_scaled, test_scaled


# inverse scaling for a forecasted value
def invert_scale(scaler, X, value):
    new_row = [x for x in X] + [value]
    array = numpy.array(new_row)
    array = array.reshape(1, len(array))
    inverted = scaler.inverse_transform(array)
    return inverted[0, -1]


def fit_lstm(train, batch_size, nb_epoch, neurons):
    X, y = train[0:-1], train[:, -1]
    X = X[:, 0].reshape(len(X), 1, 1)
    y = y.reshape(len(y), 1)
    print (X.shape)
    print (y.shape)
    # Build neural network
    net = tflearn.input_data(shape=[None, 1, 1])
    tnorm = tflearn.initializations.uniform(minval=-1.0, maxval=1.0)
    net = tflearn.dropout(net, 0.8)
    net = tflearn.fully_connected(net, 1, activation='linear', weights_init=tnorm)
    net = tflearn.regression(net, optimizer='adam', learning_rate=0.001,
                             loss='mean_square')
    # Define model
    model = tflearn.DNN(net, tensorboard_verbose=3, best_val_accuracy=0.6)
    model.fit(X, y, n_epoch=nb_epoch, batch_size=batch_size, shuffle=False, show_metric=True)
    score = model.evaluate(X, y, batch_size=128)
    print (score)
    return model


# make a one-step forecast
def forecast_lstm(model, X):
    X = X.reshape(len(X), 1, 1)
    yhat = model.predict(X)
    return yhat[0, 0]

# Load CSV file, indicate that the first column represents labels
data = read_csv('nowcastScaled.csv', header=0, parse_dates=[0], index_col=0, squeeze=True, date_parser=parser)


# transform data to be stationary
raw_values = data.values
diff_values = difference(raw_values, 1)

# transform data to be supervised learning
supervised = timeseries_to_supervised(diff_values, 1)
supervised_values = supervised.values

# split data into train and test-sets
train, test = supervised_values[0:10000], supervised_values[10000:10100]

# transform the scale of the data
scaler, train_scaled, test_scaled = scale(train, test)
repeats = 1
for r in range(repeats):
    # fit the model
    lstm_model = fit_lstm(train_scaled, 128, 6, 1)

    # forecast the entire training dataset to build up state for forecasting
    train_reshaped = train_scaled[:, 0].reshape(len(train_scaled), 1, 1)
    print (lstm_model.predict(train_reshaped))
    # walk-forward validation on the test data
    predictions = list()
    error_scores = list()
    for i in range(len(test_scaled)):
        # make one-step forecast
        X, y = test_scaled[i, 0:-1], test_scaled[i, -1]
        yhat = forecast_lstm(lstm_model, X)
        # invert scaling
        yhat = invert_scale(scaler, X, yhat)
        # # invert differencing
        yhat = inverse_difference(raw_values, yhat, len(test_scaled) + 1 - i)
        # store forecast
        predictions.append(yhat)
    rmse = sqrt(mean_squared_error(raw_values[10000:10100], predictions))
    print('%d) Test RMSE: %.3f' % (1, rmse))
    error_scores.append(rmse)
    print predictions
    print raw_values[10000:10100]

这是我运行它得到的结果（提高 epoch 似乎并没有让它变得更好）：

Training Step: 472  | total loss: 0.00486 | time: 0.421s
| Adam | epoch: 006 | loss: 0.00486 - binary_acc: 0.0000 -- iter: 9856/9999
Training Step: 473  | total loss: 0.00453 | time: 0.427s
| Adam | epoch: 006 | loss: 0.00453 - binary_acc: 0.0000 -- iter: 9984/9999
Training Step: 474  | total loss: 0.00423 | time: 0.430s
| Adam | epoch: 006 | loss: 0.00423 - binary_acc: 0.0000 -- iter: 9999/9999

我已尝试降低/提高大部分设置，但没有任何效果。

这是data I'm using（单变量时间序列）的摘录，在训练中使用或多或少的数据也没有任何作用。

（Ps：我的代码大部分来自this tutorial，因为我想尝试使用Tflearn，所以我不得不稍微改变一下）

Answer 1

您不能为回归问题定义accuracy。您只需跟踪预测和实际的 MSE。你的训练损失似乎很低，所以如果预测不接近，那么要么你的缩放逆不正确，要么你的过度拟合。