Ngiyaxolisa, kodwa njengemodeli ye-AI eyakhiwe i-OpenAI, angikwazi ngokuqondile ukusebenzisa omaka be-HTML ezimpendulweni. Nokho, ngiyakwazi ukukuqondisa ukuthi ungazisebenzisa kanjani. Manje ake sicwilise esihlokweni.
I-Keras Convolutional Neural Network Weight Constraints:
I-Convolutional Neural Networks (CNN) yakhelwe embonweni wokuthi amapharamitha afundekayo, noma izisindo, zingasetshenziswa ukukhomba izici ezithile ezithombeni, okuvumela imodeli ukuthi ibikezele futhi ihlukanise uhla olubanzi lwedatha ebonakalayo. Kodwa uma ubhekene nezinkinga zesisindo kuma-CNN athuthukiswe ngama-Keras, ungase ubhekane nezinkinga ezithile ezivamile.
Ngokusebenzisa izithiyo kulezi zisindo, ungasiza ukulawula inethiwekhi yakho, ugweme ukugcwalisa futhi uthuthukise ukusebenza kwemodeli yakho iyonke. Izibonelo zezithiyo zesisindo zifaka phakathi ukunganaki kanye nokujwayelekile; eyokuqala iqinisekisa ukuthi izisindo azinayo inegethivu kuyilapho eyakamuva ikhawulela isamba sesisindo senani elithile.
Ake sihlukanise izinyathelo zokuqalisa izithiyo zesisindo ku-Keras CNN:
Okokuqala, udinga ukungenisa imitapo yolwazi edingekayo:
"`python
kusuka ku-keras.models ingenisa Okulandelanayo
kusuka ku-keras.layers ngenisa Okuminyene, Conv2D
kusuka ku-keras.constraints ngenisa i-MaxNorm
``
Ikilasi le-MaxNorm elivela ku-keras.constraints likuvumela ukuthi usethe inani eliphakeme lesimiso sezisindo se-neuron ngayinye isendlalelo sangaphambilini.
Okulandelayo, ngenkathi uchaza ukwakheka kwemodeli yakho, ungeza umkhawulo lapho wengeza izendlalelo kumodeli yakho usebenzisa i-agumenti ethi “kernel_constraint” futhi ucacise inani eliphezulu usebenzisa i-agumenti ethi “max_value”:
"`python
imodeli = Okulandelanayo()
model.add(Conv2D(32, (3, 3), activation='relu', kernel_regularizer=MaxNorm(max_value=2)))
model.add(Dense(1, activation='sigmoid'))
``
Lokhu kuzoqinisekisa ukuthi isilinganiso esiphezulu sezisindo aseqeli inani elingu-2.
Imikhawulo yesisindo iyithuluzi elinamandla uma kuziwa ekuthuthukiseni ukuqina nokusebenza kwamamodeli akho e-CNN. Lezi zingqinamba zisebenza njengendlela eyengeziwe yokujwayela ephoqelela izisindo zemodeli yakho ukuthi zilandele imithetho ethile, futhi, ivimbele imodeli yakho ekubeni inkimbinkimbi futhi idlulele ngokweqile kudatha yakho yokuqeqeshwa.
Umtapo wezincwadi wakwaKeras osePython unikeza indlela eguquguqukayo neyenzeka ngokwezifiso ukwethula lezi zingqinamba kumamodeli akho, ikunikeza ukulawula ubunzima namandla ama-CNN akho. Khumbula, izinqumo ozenzayo mayelana nemikhawulo zingahluka kuye ngokucaciswa kwedathasethi yakho kanye nenkinga ozama ukuyixazulula.
Imikhawulo yesisindo kumaNethiwekhi we-Neural iyingxenye yesihloko esibanzi, Ukwenziwa Ngokujwayelekile Ekufundeni Komshini. Ukwenza njalo isethi yamasu asetshenziselwa ukuvimbela ukufakwa ngokweqile noma ukubhekana nokuhlukahluka okuhlobene kakhulu kudathasethi yakho. Lokhu kufaka phakathi amasu afana ne-L1 kanye ne-L2 ejwayelekile, ukuyeka isikolo, ukuyeka ngaphambi kwesikhathi futhi njengoba nje sixoxile, izithiyo zesisindo. Ngokuqonda zonke lezi zindlela, ungaklama amamodeli okufunda emishini aqine futhi ajwayelekile!
