Projekt je navrhem a implementaci neuronove site, ktera klasifikuje rukou psane cislice. Jako trenovaci a testovaci sada byly pouzite vzorky z MNIST databaze, ktera je podmnouzinou vetsi databaze NIST. Data se skladaji z 60000 trenovacich dat a 10000 testovacich dat. Databaze je dostupna zdarma online a je nutne si ji stahnout pro spusteni projektu. MNIST
Dosazene vysledky je mozne porovnat s jinymi implementacemi, ktere na stejnem miste publikovane.
Pro beh aplikace je nutne mit nainstalovany Java Runtime Environment ve verzi alespon 1.7
Projekt je mozne spustit nasledujicim prikazem :
java -cp target/NeuralNetwork-1.0-SNAPSHOT.jar \
fit.NNSFC.xkidon00.NeuralNetwork \
data/train-images-idx3-ubyte \
data/train-labels-idx1-ubyte \
data/t10k-images-idx3-ubyte \
data/t10k-labels-idx1-ubyte
kde jednotlive parametry:
| Parametr | Vyznam |
|---|---|
target/NeuralNetwork-1.0-SNAPSHOT.jar |
archiv s projektem |
fit.NNSFC.xkidon00.NeuralNetwork |
trida ktera se ma spustit |
data/train-images-idx3-ubyte |
trenovaci data |
data/train-labels-idx1-ubyte |
ohodnoceni trenovacich dat |
data/t10k-images-idx3-ubyte |
testovaci data |
data/t10k-labels-idx1-ubyte |
ohodnoceni testovacich dat |
after batch : 0 out of : 60000 -> correctly classified : 1069 out of : 10000
after batch : 60 out of : 60000 -> correctly classified : 1005 out of : 10000
after batch : 120 out of : 60000 -> correctly classified : 1025 out of : 10000
after batch : 180 out of : 60000 -> correctly classified : 1162 out of : 10000
after batch : 240 out of : 60000 -> correctly classified : 1563 out of : 10000
after batch : 300 out of : 60000 -> correctly classified : 1735 out of : 10000
.
.
.
after batch : 4380 out of : 60000 -> correctly classified : 2740 out of : 10000
after batch : 4440 out of : 60000 -> correctly classified : 2779 out of : 10000
after batch : 4500 out of : 60000 -> correctly classified : 2806 out of : 10000
after batch : 4560 out of : 60000 -> correctly classified : 2814 out of : 10000
after batch : 4620 out of : 60000 -> correctly classified : 2835 out of : 10000
after batch : 4680 out of : 60000 -> correctly classified : 2833 out of : 10000