Koyarwar TensorFlow ga Masu Farawa
TensorFlow yana ɗaya daga cikin shahararrun tsarin ilmantarwa mai zurfi da koyon injina. An haɓaka TensorFlow ta ƙungiyar Google Brain, kuma an yi amfani da shi sosai a cikin ayyukan bincike da aikace-aikacen masana'antu da yawa. Wannan labarin yana ba da koyaswa mataki-mataki don taimaka muku, a matsayinku na mafari, fara da TensorFlow.
1. Fahimtar Muhimmancin TensorFlow
Kafin mu fara shigarwa da amfani da TensorFlow, yana da mahimmanci a fahimci menene TensorFlow da kuma mahimman ra'ayoyin da ke bayansa. TensorFlow tsarin buɗe tushen bayanai ne don lissafin lambobi da koyon injin. Yana amfani da jadawalin kwararar bayanai don yin ayyukan lambobi, inda nodes a cikin jadawalin ke wakiltar ayyukan lissafi, kuma gefuna suna wakiltar jerin bayanai masu girma dabam dabam (tensors) da aka haɗa a tsakaninsu.
2. Shigar da TensorFlow
Mataki na farko wajen amfani da TensorFlow shine shigar da shi. Ga yadda ake shigar da TensorFlow ta amfani da pip, manajan kunshin Python.
1. Shigar da Python:
Tabbatar kana da Python a tsarinka. TensorFlow ya dace da Python 3.6 zuwa 3.9 a lokacin rubuta wannan rubutun. Kuna iya saukar da Python daga gidan yanar gizon Python na hukuma.
2. Muhalli na Zamani:
Ana ba da shawarar sosai don ƙirƙirar yanayi na kama-da-wane don ware aikin TensorFlow ɗinku:
"`sh
python -m venv myenv
tushen myenv/bin/kunnawa Ga masu amfani da Mac/Linux
myenv\Scripts\activate Ga masu amfani da Windows
““
3. Shigar da TensorFlow:
Yanzu, shigar da TensorFlow ta amfani da pip:
"`sh
pip shigar tensorflow
““
3. Sannu Duniya tare da TensorFlow
Yanzu da aka shigar da TensorFlow, bari mu ƙirƙiri wani rubutun Python mai sauƙi don tabbatar da shigarwar. Ƙirƙiri sabon fayil ɗin Python kuma a sanya masa suna `hello_tensorflow.py`.
"' Python
shigo da tensorflow kamar tf
Ƙirƙiri abin da ke canzawa
sannu = tf.constant('Sannu, TensorFlow!')
Fara zaman
tare da tf.Session() a matsayin sess:
sakamako = sess.run (sannu)
buga(sakamako)
““
Daidaita lambar bisa ga sigar TensorFlow 2.x:
"' Python
shigo da tensorflow kamar tf
Ƙirƙiri abin da ke canzawa
sannu = tf.constant('Sannu, TensorFlow!')
Gudu ta amfani da aiwatar da sha'awa (a kunna ta tsohuwa)
buga(sannu.numpy())
““
Ajiye fayil ɗin, sannan a gudanar da:
"`sh
Python hello_tensorflow.py
““
4. Fahimtar Tensors da Ayyukan Asali
Tensors sune babban tsarin bayanai a cikin TensorFlow, waɗanda sune jeri mai girma dabam-dabam. Ga wasu misalai don taimaka muku fahimtar tensors:
"' Python
shigo da tensorflow kamar tf
Ƙirƙirar tensors
scalar = tf. madaidaici (7) scalar
vector = tf. constant([1, 2, 3]) vector
matrix = tf. constant([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) tensor na 3D
buga(f'Scalar: {scalar}')
buga(f'Vector: {vector}')
buga(f'Matrix: {matrix}')
buga(f'Tensor 3D: {tensor3d}')
““
Don yin ayyukan asali akan tensors:
"' Python
a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 6], [7, 8]])
Aikin ƙari
ƙara = tf.ƙara(a, b)
Ayyukan ninka matrix
mul = tf.matmul(a, b)
buga(f'Ƙari: {ƙara}')
buga(f'Matrix Multiplication: {mul}')
““
5. Ƙirƙirar Tsarin Sadarwa Mai Sauƙi na Jijiyoyi
Mataki na gaba shine ƙirƙirar samfurin hanyar sadarwa mai sauƙi. Za mu gina samfurin rarraba hotuna ta amfani da bayanan MNIST, wani rumbun adana bayanai na hotunan lambobi da aka rubuta da hannu. Bari mu fara:
"' Python
shigo da tensorflow kamar tf
daga tensorflow.keras shigo da bayanai, yadudduka, samfura
Sauke bayanan MNIST
(hotunan_jirgin_kafa, alamun_jirgin_kafa), (hotunan_jirgin_kafa, alamun_jirgin_kafa) = bayanai.mnist.bayanan_jirgin_kafa()
Daidaita hoto
hotunan_jirgin ƙasa, hotunan_jirgin ƙasa = hotunan_jirgin ƙasa / 255.0, hotunan_jirgin ƙasa / 255.0
Yin samfuri
samfurin = samfura. Jerin abubuwa([
yadudduka. Faɗaɗa(siffar_shigar=(28, 28)),
Layers.Dense(128, kunnawa='relu'),
yadudduka.Masu kauri(10)
])
Tarin samfura
model.compile(optimizer='adam',
asara=tf.keras.asarar.SparseClassicalCrosentropy(daga_logits=Gaskiya),
awo = ['daidai']))
Horar da samfurin
model.daidaitacce(hotunan_jirgin_kafa, alamun_jirgin_kafa, lokutan zamani=5)
Gwada samfurin
gwajin_asarar, gwajin_acc = samfuri. kimantawa (hotunan_gwaji, alamun_gwaji)
buga(f'Gwajin daidaito: {gwajin_acc}')
““
Bayani:
– Saitin Bayanai: Muna shigo da bayanai na MNIST kuma muna loda su.
- Tsarin sarrafawa: Daidaita bayanai ta hanyar raba ƙimar pixel da 255.
– Samfuri: Mun ayyana samfuri mai sauƙi mai layuka biyu. Mataki na farko shine Layer 'Flatten' don canza hoton 2D zuwa jerin 1D. Mataki na biyu shine Layer 'Dense' mai ƙwayoyin jijiyoyi 128 da 'relu' a matsayin aikin kunnawa, kuma na ƙarshe shine Layer 'Dense' mai ƙwayoyin jijiyoyi 10 waɗanda ke wakiltar azuzuwan 10.
– Tattara: Muna tattara samfurin ta amfani da ma'aunin "adam" da kuma "SparseCategoricalCrossentropy" a matsayin aikin asara.
– Horarwa: Horar da samfurin na tsawon lokaci 5.
– Kimantawa: Kimanta samfurin bisa ga bayanan gwaji.
6. Ajiyewa da Loda Samfura
Bayan horar da samfurin, za ka iya son adana shi don amfani daga baya ba tare da sake horar da shi ba. Ga yadda ake adanawa da loda samfurin:
"' Python
Ajiye samfurin
samfurin.save('my_model.h5')
Tsarin lodawa
sabon_samfuri = tf.keras.models.load_model('my_model.h5′)
Tabbatar da samfurin da aka ɗora
asara, acc = sabon_samfuri. kimantawa(hotunan_gwaji, alamun_gwaji)
bugawa(f'Daidaicin samfurin da aka ɗora: {acc}')
““
Kammalawa
Wannan jagorar tana ba da cikakken bayani game da fara amfani da TensorFlow ga masu farawa. Mun rufe shigarwa, ayyukan tensor na asali, da gina samfurin hanyar sadarwa mai sauƙi ta amfani da bayanan MNIST. TensorFlow yana ba da damar bincike da yawa, kamar sarrafa bayanai na ci gaba, samfuran da suka fi rikitarwa, da amfani da TensorFlow akan na'urori kamar TPUs da GPUs. Muna fatan wannan koyaswar za ta taimaka muku fara a duniyar koyon injina tare da TensorFlow.
