neuroflow.playground.DigitRecognition.scala Maven / Gradle / Ivy
package neuroflow.playground
import breeze.linalg.DenseVector
import neuroflow.application.plugin.Extensions._
import neuroflow.application.plugin.IO.Jvm._
import neuroflow.application.plugin.Notation._
import neuroflow.application.processor.Image._
import neuroflow.common.~>
import neuroflow.core
import neuroflow.core.Activators.Float._
import neuroflow.core._
import neuroflow.dsl._
import neuroflow.nets.cpu.DenseNetwork._
/**
* @author bogdanski
* @since 04.01.16
*/
object DigitRecognition {
/*
Here the goal is to classify digits from unknown font family 'h'.
Feel free to read this article for the full story:
http://znctr.com/blog/digit-recognition
*/
def digitSet2Vec(path: String): Seq[DenseVector[Float]] = {
val selector: Int => Boolean = _ < 255
(0 to 9) map (i => loadBinary(getResourceFile(path + s"$i.png"), selector).float)
}
def apply = {
val sets = ('a' to 'h') map (c => digitSet2Vec(s"img/digits/$c/"))
val xs = sets.dropRight(1).flatMap { s => (0 to 9).map { digit => s(digit) } }
val ys = sets.dropRight(1).flatMap { m =>
(0 to 9).map { digit =>
val t = zero[Float](10)
t.update(digit, 1.0f)
t
}
}
val config = (1 to 3).map(_ -> (0.0f, 0.01f)) :+ 4 -> (0.0f, 0.1f)
implicit val weights = WeightBreeder[Float].normal(config.toMap)
val (f, g) = (ReLU.biased(0.1f), ReLU.biased(1.0f))
val L =
Vector (xs.head.length) ::
Dense (400, f) ::
Dense (200, f) ::
Dense (50, f) ::
Dense (10, g) :: SoftmaxLogMultEntropy[Float](N = 1)
val net = Network(
layout = L,
settings = Settings[Float](
learningRate = { case (_, _) => 1E-4f },
prettyPrint = true,
updateRule = Momentum(0.8f),
precision = 1E-3,
iterations = 15000
)
)
net.train(xs, ys)
val posWeights = net.weights.foldLeft(0)((count, m) => count + m.findAll(_ > 0.0).size)
val negWeights = net.weights.foldLeft(0)((count, m) => count + m.findAll(_ < 0.0).size)
println(s"Pos: $posWeights, Neg: $negWeights")
val setsResult = sets.map(s => s.map(v => net(v)))
('a' to 'h') zip setsResult foreach {
case (char, res) =>
~> (println(s"set $char:")) next (0 to 9) foreach { digit =>
println(s"$digit classified as " + res(digit).toScalaVector.indexOf(res(digit).max))
}
}
}
}
/*
_ __ ________
/ | / /__ __ ___________ / ____/ /___ _ __
/ |/ / _ \/ / / / ___/ __ \/ /_ / / __ \ | /| / /
/ /| / __/ /_/ / / / /_/ / __/ / / /_/ / |/ |/ /
/_/ |_/\___/\__,_/_/ \____/_/ /_/\____/|__/|__/
1.7.0
Network : neuroflow.nets.cpu.DenseNetwork
Weights : 170.500 (≈ 0,650406 MB)
Precision : Single
Loss : neuroflow.core.SoftmaxLogEntropy
Update : neuroflow.core.Momentum
Layout : 200 Vector
400 Dense (ReLU + Bias(0.1))
200 Dense (ReLU + Bias(0.1))
50 Dense (ReLU + Bias(0.1))
10 Dense (ReLU + Bias(1.0))
O
O
O O O
O O O
O O O O O
O O O O O
O O O
O O O
O
O
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:18:887] Training with 70 samples, batch size = 70, batches = 1.
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:19:259] Breeding batches ...
[scala-execution-context-global-67] DEBUG neuroflow.core.BatchBreeder$ - [13.08.2018 20:58:19:396] Bred Batch 0.
Aug 13, 2018 8:58:19 PM com.github.fommil.jni.JniLoader liberalLoad
INFORMATION: successfully loaded /var/folders/t_/plj660gn6ps0546vj6xtx92m0000gn/T/jniloader1733496608378989451netlib-native_system-osx-x86_64.jnilib
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:19:536] Iteration 1.1, Avg. Loss = 16,1312, Vector: 15.695352 16.388042 15.79654 16.736338 16.023125 17.038774 ... (10 total)
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:19:559] Iteration 2.1, Avg. Loss = 16,1311, Vector: 15.695781 16.387875 15.796873 16.735378 16.02314 17.037626 ... (10 total)
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:19:569] Iteration 3.1, Avg. Loss = 16,1310, Vector: 15.696555 16.387573 15.797471 16.733648 16.023165 ... (10 total)
...
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:36:281] Iteration 2513.1, Avg. Loss = 0,0100014, Vector: 0.009870188 0.012217878 0.016114127 0.0054787886 ... (10 total)
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:36:288] Iteration 2514.1, Avg. Loss = 0,00999423, Vector: 0.0098633645 0.012208382 0.01610323 0.0054752673 ... (10 total)
[run-main-0] INFO neuroflow.nets.cpu.DenseNetworkFloat - [13.08.2018 20:58:36:288] Took 2514 of 15000 iterations.
Pos: 100289, Neg: 70211
set a:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set b:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set c:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set d:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set e:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set f:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set g:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
set h:
0 classified as 0
1 classified as 1
2 classified as 2
3 classified as 3
4 classified as 4
5 classified as 5
6 classified as 6
7 classified as 7
8 classified as 8
9 classified as 9
[success] Total time: 21 s, completed 13.08.2018 20:58:36
*/© 2015 - 2025 Weber Informatics LLC | Privacy Policy