def geneateRandomWeight(shape: Int*): Tensor = 2 * np.rand(shape.toArray) - 1

val biasedTrainSet = np.concatenate(Seq(trainSet, np.ones(trainSet.shape(0)).T), 1)

def initializeWeights(layers: List[Int], weights: List[Tensor] = List()): List[Tensor] = {

    // the last layer is our networks target, so doesn't need any further weights
    if(layers.length < 2) return weights

    // the second last layer doesn't need to take an account an additional bias column
    if(layers.length == 2) weights :+ geneateRandomWeight(layers.head + 1, layers.last)

    // generate weigths aware of added bias column
    else initializeWeights(layers.tail, weights :+ geneateRandomWeight(layers.head + 1, layers.tail.head + 1))
  }