/// <summary>
/// Get the hidden layer features from a visible layer
/// ---------
/// Assuming the RBM has been trained (so that weights for the network have been learned),
/// run the network on a set of hidden units, to get a sample of the visible units.
///
/// Parameters
/// ---------
/// data: A matrix where each row consists of the states of the hidden units.
///
/// Returns
/// ---------
/// visible_states: A matrix where each row consists of the visible units activated from the hidden
/// units in the data matrix passed in.
/// </summary>
public double[][] GetHiddenLayer(double[][] dataArray)
{
var num_examples = dataArray.Length;
// Create a matrix, where each row is to be the hidden units (plus a bias unit)
// sampled from a training example.
var hidden_states = RealMatrix.Ones(num_examples, m_numHiddenElements + 1);
var data = new RealMatrix(dataArray);
// Insert bias units of 1 into the first column of data.
data = data.InsertCol(1); // np.insert(data, 0, 1, axis = 1)
// Calculate the activations of the hidden units.
var hiddenActivations = data * m_weights;
// Calculate the probabilities of turning the hidden units on.
//激活函数是指的如何把“激活的神经元的特征”通过函数把特征保留并映射出来，这是神经网络能解决非线性问题关键
//激活函数，就是在神经网络的神经元上运行的函数，负责将神经元的输入映射到输出端。
var hiddenProbs = ActivationFunctions.Logistic(hiddenActivations);
// Turn the hidden units on with their specified probabilities.
//以指定的概率打开隐藏的单元
hidden_states = hiddenProbs > Distributions.UniformRandromMatrix(num_examples, m_numHiddenElements + 1);
// Ignore the bias units.
hidden_states = hidden_states.RemoveFirstCol();
return(hidden_states);
}
/// <summary>
/// Day dream - Reconstruct a randrom matrix (An interesting way of seeing strong features the machine has learnt).
///
/// Randomly initialize the visible units once, and start running alternating Gibbs sampling steps
/// (where each step consists of updating all the hidden units, and then updating all of the visible units),
/// taking a sample of the visible units at each step.
///
/// Note that we only initialize the network "once", so these samples are correlated.
/// ---------
/// samples: A matrix, where each row is a sample of the visible units produced while the network was daydreaming。
/// </summary>
/// <param name="numberOfSamples">How many images/dreams</param>
/// <returns>Array of Reconstructed dreams</returns>
public double[][] DayDream(int numberOfSamples)
{
//Create a matrix, where each row is to be a sample of of the visible units
//(with an extra bias unit), initialized to all ones.
var data = RealMatrix.Ones(numberOfSamples, m_numVisibleElements + 1);
//Take the first sample from a uniform distribution.
data.Update(0, 1, Distributions.UniformRandromMatrixBool(1, m_numVisibleElements), 1);
//Start the alternating Gibbs sampling.
//Note that we keep the hidden units binary states, but leave the
//visible units as real probabilities.
//See section 3 of Hinton's "A Practical Guide to Training Restricted Boltzmann Machines" for more on why.
for (int i = 0; i < numberOfSamples; i++)
{
var visible = data.Submatrix(i, 0, 1).ToVector();
//Calculate the activations of the hidden units.
var hidden_activations = (visible * m_weights).ToVector();
//Calculate the probabilities of turning the hidden units on.
var hidden_probs = ActivationFunctions.Logistic(hidden_activations);
//Turn the hidden units on with their specified probabilities.
var hidden_states = hidden_probs > RVector.Random(m_numHiddenElements + 1);
//Always fix the bias unit to 1.
hidden_states[0] = 1;
//Recalculate the probabilities that the visible units are on.
var visible_activations = (hidden_states * m_weights.Transpose).ToVector();
var visible_probs = ActivationFunctions.Logistic(visible_activations);
var visible_states = visible_probs > RVector.Random(m_numVisibleElements + 1);
data.Update(visible_states, 0, false, i, 0);
}
return(data.Submatrix(0, 1).ToArray());
}
/// <summary>
/// Get the visible layer from a hidden layer
/// ---------
/// Assuming the RBM has been trained (so that weights for the network have been learned),
/// run the network on a set of visible units, to get a sample of the hidden units.
/// Parameters
/// ----------
/// data: A matrix where each row consists of the states of the visible units.
///
/// Returns
/// -------
/// hidden_states: A matrix where each row consists of the hidden units activated from the visible
/// units in the data matrix passed in.
/// </summary>
public double[][] GetVisibleLayer(double[][] dataArray)
{
var numExamples = dataArray.Length;
// Create a matrix, where each row is to be the visible units (plus a bias unit)
// sampled from a training example.
var data = new RealMatrix(dataArray);
// Insert bias units of 1 into the first column of data.
data = data.InsertCol(1);
// Calculate the activations of the visible units.
var visibleActivations = data * m_weights.Transpose;
// Calculate the probabilities of turning the visible units on.
var visibleProbs = ActivationFunctions.Logistic(visibleActivations);
// Turn the visible units on with their specified probabilities.
var visibleStates = visibleProbs > Distributions.UniformRandromMatrix(numExamples, m_numVisibleElements + 1);
// Always fix the bias unit to 1
// Ignore the bias units.
visibleStates = visibleStates.RemoveFirstCol(); //visible_states[:,1:]
return(visibleStates);
}
//public void AsyncTrain(double[][] data, int maxEpochs)
//{
// double e = 0;
// var f = new TaskFactory();
// f.StartNew(new Action(() => Train(data, maxEpochs, out e)));
//}
//原理参考：http://www.cnblogs.com/pinard/p/6530523.html或者http://deeplearning.net/tutorial/rbm.html
public void Train(double[][] dataArray, int maxEpochs, out double error)
{
error = 0;
var numExamples = dataArray.Length;
var data = new RealMatrix(dataArray);
// Insert bias units of 1 into the first column.
data = data.InsertCol(1);
Stopwatch sw = new Stopwatch();//准确地测量运行时间。
for (int i = 0; i < maxEpochs; i++)
{
sw.Start();
//Clamp to the data and sample from the hidden units.
//This is the "positive CD phase", aka the reality phase
var posHiddenActivations = data * m_weights;
var posHiddenProbs = ActivationFunctions.Logistic(posHiddenActivations); //隐层神经元被激活的概率
posHiddenProbs = posHiddenProbs.Update(0, 1, 1); // 相当于P(hj=1|v)，Fix the bias unit
var posHiddenStates = posHiddenProbs > Distributions.UniformRandromMatrix(numExamples, m_numHiddenElements + 1);
// Note that we're using the activation "probabilities" of the hidden states, not the hidden states
// themselves, when computing associations. We could also use the states; see section 3 of Hinton's
// "A Practical Guide to Training Restricted Boltzmann Machines" for more
var posAssociations = data.Transpose * posHiddenProbs;
// Reconstruct the visible units and sample again from the hidden units
// This is the "negative CD phase", aka the daydreaming phase
var negVisibleActivations = posHiddenStates * m_weights.Transpose;
var negVisibleProbs = ActivationFunctions.Logistic(negVisibleActivations); //隐藏层到可见层
negVisibleProbs = negVisibleProbs.Update(0, 1, 1); //相当于P(vj=1|h)
var negHiddenActivations = negVisibleProbs * m_weights;
var negHiddenProbs = ActivationFunctions.Logistic(negHiddenActivations);
// Note, again, that we're using the activation "probabilities" when computing associations, not the states themselves
var negAssociations = negVisibleProbs.Transpose * negHiddenProbs;
// Update weights
m_weights = m_weights + (m_learningRate * ((posAssociations - negAssociations) / numExamples));
sw.Stop(); //计时结束
error = (((data - negVisibleProbs) ^ 2).Average()) / dataArray.Length; //误差除以size
RaiseEpochEnd(i, error); //记录每次学习的误差
Console.WriteLine("Epoch {0}: error is {1}, computation time (ms): {2}", i, error, sw.ElapsedMilliseconds);
sw.Reset(); //运行时间sw清零
}
RaiseTrainEnd(maxEpochs, error);
}
