func (self *Sequential) Train(data [][]float64, labels [][]float64, epochs int, batchSize int) {
//
fmt.Println("\n\n *** TRAINING...")
// Set the regularisation parameter.
L2REG := 0.0
// Train for all epochs.
for e := 0; e < epochs; e++ {
// Create a random permutation of the data.
randomPerm := rand.Perm(len(data))
//
error := 0.0
accuracy := 0.0
// Loop over the dataset.
for d := 0; d < len(data); d++ {
self.FeedForward(data[randomPerm[d]])
error += self.CalculateCost(labels[randomPerm[d]], L2REG)
self.BackPropagate(labels[randomPerm[d]], batchSize, (d+1)%batchSize == 0, L2REG)
if utils.Argmax(self.layers[self.numLayers-1].Activations) == utils.Argmax(labels[randomPerm[d]]) {
accuracy++
}
if (d+1)%500 == 0 {
fmt.Printf("\tEpoch %vth / Progress: %.2f%% / Avg. Loss: %.4f / Accuracy %.2f%% \r", e+1, float64(d+1)*100/float64(len(data)), error/float64(d+1), 100.0*accuracy/float64(d+1))
}
}
fmt.Printf("\r\n")
}
}
func (self *Sequential) Predict(data [][]float64, labels [][]float64) {
fmt.Println(" *** PREDICTING TEST-DATA")
accuracy := 0.0
for d := 0; d < len(data); d++ {
self.FeedForward(data[d])
if utils.Argmax(self.layers[self.numLayers-1].Activations) == utils.Argmax(labels[d]) {
accuracy++
}
if (d+1)%100 == 0 {
fmt.Printf("\tProgress: %.2f%% / Accuracy %.2f%% \r", float64(d+1)*100/float64(len(data)), 100.0*accuracy/float64(d+1))
}
}
}