An warware: lodin samfurin pytorchand daskare

A cikin duniyar yau, inganta aikin ƙirar injina ya zama muhimmin aiki ga masu haɓakawa da masana kimiyyar bayanai. Wata hanyar da aka saba amfani da ita don yin haka ita ce ta hanyar amfani da dabarun "samfurin kaya" da "daskare" dabaru. A cikin wannan labarin, za mu tattauna yadda waɗannan hanyoyin ke taimakawa wajen inganta samfurin, yadda ake aiwatar da su a Python, da wasu muhimman al'amura da suka shafi matsalar da aiwatarwa.

Load model da kuma daskare dabaru ne guda biyu waɗanda za a iya amfani da su don haɓaka aiki da inganci na ƙirar koyon injin. Na farko ya ƙunshi loda samfurin da aka riga aka horar don yin amfani da fasalulluka maimakon horar da sabon ƙira daga karce, yayin da na ƙarshen yana nufin dakatar da sabunta wasu ma'aunin nauyi yayin aikin horo don tacewa da haɓaka aikin ƙirar. Dukansu fasahohin suna taimakawa wajen rage wuce gona da iri kuma suna iya taimakawa gina ingantattun samfura masu inganci.

Aiwatar da Model Load da Daskare a Python

Domin aiwatar da samfurin lodi da kuma daskare dabarun yadda ya kamata, za mu fara buƙatar samun samfurin da aka riga aka horar a hannunmu. Don wannan misalin, za mu yi amfani da Python tare da shahararrun ɗakunan karatu na koyon injin kamar TensorFlow da Keras don nuna matakan.

import tensorflow as tf
from tensorflow.keras import layers

# Load a pre-trained model
model = tf.keras.applications.VGG16(weights='imagenet', include_top=False)

# Set specific layers as non-trainable (frozen)
for layer in model.layers[:10]:
    layer.trainable = False

# Add custom layers on top of the pre-trained model
x = model.output
x = layers.GlobalAveragePooling2D()(x)
x = layers.Dense(1024, activation='relu')(x)
predictions = layers.Dense(10, activation='softmax')(x)

# Finalize the new model
custom_model = tf.keras.Model(inputs=model.input, outputs=predictions)

Ana Load da Samfuran da aka riga aka horar

The kaya model tsari yana farawa ta hanyar shigo da samfurin da aka riga aka horar, kamar VGG16, wanda aka horar da shi akan bayanan bayanan ImageNet. TensorFlow da Keras suna ba da madaidaiciyar hanyoyi don shigo da irin waɗannan samfuran, kamar yadda aka gani a lambar da ke sama. Amfanin yin amfani da samfurin da aka riga aka horar shi ne cewa ya riga ya koyi abubuwan da suka dace daga babban bayanan bayanai, yana ba mu damar yin amfani da wannan ilimin yayin horar da tsarin mu na al'ada, da rage yawan lokaci da albarkatun lissafi.

Daskarewar Yadudduka da Ƙara Yadudduka na Musamman

Da zarar an ɗora samfurin da aka riga aka horar, za mu iya ci gaba zuwa daskare takamaiman yadudduka na samfurin don hana su daga sabunta su yayin horo. A cikin wannan misalin, mun daskare yadudduka 10 na farko na samfurin VGG16, muna saita sifa ta "masu horo" zuwa Ƙarya. Daskare waɗannan yadudduka yana ba samfurin damar riƙe abubuwan da aka koya a baya da kuma mai da hankali kan tace yadudduka na gaba don ingantaccen aiki.

Bayan daskarewa da yadudduka da ake so, sai mu ƙara yadudduka na al'ada a saman samfurin da aka riga aka horar bisa ga bukatunmu. Aiwatar da mu tana nuna ƙari na GlobalAveragePooling2D Layer wanda ke biye da yadudduka masu yawa waɗanda ke aiki azaman yadudduka na fitarwa don ƙirar mu ta al'ada. A ƙarshe, muna haɗa samfurin da aka riga aka horar da tsarin ƙirar mu na al'ada a cikin sabon ƙirar ta amfani da hanyar tf.keras.Model.

Ta hanyar amfani da samfurin lodi da daskare dabaru tare da Python, TensorFlow, da Keras, mun sami nasarar inganta aikin ƙirar mu. Wannan haɗin kayan aiki masu ƙarfi da dabaru za su ba wa masana kimiyyar bayanai da masu haɓakawa damar ƙirƙirar ingantattun samfuran koyo na inji waɗanda suke daidai kuma masu dacewa da albarkatu.