Pokud 3x provedeme nov?? (u??en?? od 0) u??en?? je mo??n?? ??e bude m??t poka??d?? jin?? vysledky (synaptick?? v??hy jsou na za????tku nastaveny n??hodn?? mezi 0 a 1)
    
Aby byla neuronov?? s??t p??i klasifikaci co nej??sp????n??j???? je nutn?? u??en?? zopakovat mnohokr??t po sob?? (Klidn?? se stejn??my obr??zky)
        
pro ????el ke kter??mu neuronovou s???? pou????v??m by bylo vhodn?? zapokavat u????c?? sadu alespo?? 50 000
        
Bohu??el dephi nen?? p????li?? jazykem rychl??m a tolik opakov??n?? nen?? ??asov?? p????pustn??
        
Z tohoto d??vodu soubor p??ilo??en?? na disku (Uk??zka synaptick??ch vah po u??en?? - pam???? neuronov?? s??t??) m?? za sebou pouze 4 000 opakov??n??
        
toto je d??vod pro?? neuronav?? s???? nejsp????e spoustu nezn??m??ch obr??zk?? (Nebyly v u????c?? sad??) za??ad?? ??patn??

Spus??te aplikaci
    
menu>soubor>otev????t soubor pam??ti...
    
vyberte soubor pam??ti
    
menu>mo??nosti>adres???? s obr??zky...
    
vyberte adres???? kter?? chcete klasifikovat
    
Nyn?? se provede klasifikace obr??zk??
    
Klikn??te na kategorii ze kter?? obr??zky chcete zobrazit

Spus??te aplikaci

menu>soubor>spustit u??en?? ze slo??ky...

vyberte adres???? kter?? obashuje u????c?? sadu (40 obr??zk??) 0..9 -> interi??r, 10..19 -> scan, 20..29 -> no??n?? krajina, 30..39 -> krajina

menu>mo??nosti>adres???? s obr??zky...
    Pozn??mka: U??en?? nep??inese p????li?? zlep??en??, proto??e tr??novac?? sadu si spust?? pouze 101 (Lze zm??nit v k??du jedn?? se o konstantu) d??ky tomu v??t??inu obr??zk?? neza??ad?? spr??vn??, nicm??n?? by ji?? m??la b??t schopn?? n??jak?? obr??zek/obr??zky za??adit spr??vn??

vyberte adres???? kter?? chcete klasifikovat

Nyn?? se provede klasifikace obr??zk??

Klikn??te na kategorii ze kter?? obr??zky chcete zobrazit

learnLoop ... Po??et opakov??n?? u????c?? sady

numOfNeuronInHiddenLayer ... po??et neuron?? + 1 v skryt??ch vrstev??ch neuronov?? s??t??. (????m v??t???? t??m v??ce se neuronov?? s??t zam??ruje na detaily)

learnspeed ... rychlost u??en?? ovliv??uje velikost korekc?? synaptick??ch vah (Zmen??uje/Zv??t??uje) 1 je norm??ln?? rychlost

If we do 3 times (learning from 0) learning, it is possible that they will have different results each time (synaptic scales are at random set at random between 0 and 1)

In order for the neural network to be the most successful in classifying, it is necessary to repeat the teachings many times in 

succession (Quietly with the same pictures)
????????????????
for the purpose to which I am using the neuron network, it would be advisable to engage a learning set of at least 50,000
????????????????
Unfortunately, dephi is not a fast language, and so many repetitions are not time-consuming
????????????????
For this reason, the file attached to the disk (Synaptic Weighing Sample - Neural Network Memory) has only 4,000 reps
????????????????
this is why the neural network probably has a lot of unknown images (They were not in the learning set)

Run the app
????????????????
menu> file> open memory file ...
????????????????
select a memory file
????????????????
menu> options> folder with pictures ...
????????????????
select the directory you want to classify

Image classification is now performed
    
Click on the category from which images you want to view

Run the app
    
menu> file> start learning from folder ... (40 pictures) 0..9 -> interior, 10..19 -> scan, 20..29 -> night landscape, 30..39 -> landscape
    
menu> options> folder with pictures ...
    
Note: Learning will not bring any improvement, because the training set will only start 101 (It can be changed in the code is a constant), because most of the pictures do not perform correctly, but some picture should be able