Maphunziro a TensorFlow kwa Oyamba
TensorFlow ndi imodzi mwa njira zodziwika bwino zophunzirira mozama komanso kuphunzira makina. Yopangidwa ndi gulu la Google Brain, TensorFlow yagwiritsidwa ntchito kwambiri m'mapulojekiti ambiri ofufuza komanso ntchito zamafakitale. Nkhaniyi ikupereka phunziro la sitepe ndi sitepe kuti likuthandizeni, monga woyamba, kuyamba ndi TensorFlow.
1. Kumvetsetsa Zoyambira za TensorFlow
Tisanayambe kukhazikitsa ndikugwiritsa ntchito TensorFlow, ndikofunikira kumvetsetsa zomwe TensorFlow ili ndi mfundo zoyambira zomwe zili kumbuyo kwake. TensorFlow ndi njira yotseguka yowerengera manambala ndi kuphunzira kwa makina. Imagwiritsa ntchito ma graph oyenda deta kuti igwire ntchito zamanambala, pomwe ma node mu graph amayimira ntchito zamasamu, ndipo m'mphepete mwake mumayimira ma multidimensional data arrays (ma tensors) olumikizidwa pakati pawo.
2. Kukhazikitsa kwa TensorFlow
Gawo loyamba logwiritsa ntchito TensorFlow ndikuyiyika. Umu ndi momwe mungayikitsire TensorFlow pogwiritsa ntchito pip, Python package manager.
1. Kukhazikitsa kwa Python:
Onetsetsani kuti mwayika Python pamakina anu. TensorFlow imagwirizana ndi Python 3.6 mpaka 3.9 panthawi yolemba izi. Mutha kutsitsa Python kuchokera patsamba lovomerezeka la Python.
2. Malo Ochitira Zinthu Pa Intaneti:
Ndikofunikira kwambiri kupanga malo enieni kuti mulekanitse pulojekiti yanu ya TensorFlow:
“`sh
python -m venv myenv
source myenv/bin/activate Kwa ogwiritsa ntchito Mac/Linux
myenv\Scripts\activate Kwa ogwiritsa ntchito Windows
``
3. Kukhazikitsa kwa TensorFlow:
Tsopano, ikani TensorFlow pogwiritsa ntchito bomba:
“`sh
kukhazikitsa tensorflow pa pip
``
3. Moni Dziko ndi TensorFlow
Tsopano popeza TensorFlow yayikidwa, tiyeni tipange script yosavuta ya Python kuti titsimikizire kuyika. Pangani fayilo yatsopano ya Python ndikuyitcha `hello_tensorflow.py`.
"`python
import tensorflow ngati tf
Pangani chosasintha
moni = tf.constant('Moni, TensorFlow!')
Yambani gawo
ndi tf.Session() ngati sess:
zotsatira = sess.run (moni)
sindikiza (zotsatira)
``
Sinthani khodi molingana ndi TensorFlow version 2.x:
"`python
import tensorflow ngati tf
Pangani chosasintha
moni = tf.constant('Moni, TensorFlow!')
Gwiritsani ntchito pulogalamu ya 'eager execution' (yomwe imachitika nthawi zonse)
sindikizani (moni.numpy())
``
Sungani fayilo, kenako thamangani:
“`sh
python hello_tensorflow.py
``
4. Kumvetsetsa Ma Tensor ndi Ntchito Zoyambira
Ma Tensor ndiye kapangidwe ka deta koyambirira mu TensorFlow, komwe ndi ma multidimensional arrays. Nazi zitsanzo zina zomwe zingakuthandizeni kumvetsetsa ma tensor:
"`python
import tensorflow ngati tf
Kupanga ma tensor
scalar = tf. chosasintha(7) scalar
vekitala = tf. chosasintha([1, 2, 3]) vekitala
matrix = tf. nthawi zonse([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) tensor ya 3D
print(f'Scalar: {scalar}')
sindikizani (f'Vekitala: {vekitala}')
sindikizani(f'Matrix: {matrix}')
sindikizani (f'Tensor 3D: {tensor3d}')
``
Kuchita ntchito zoyambira pa ma tensor:
"`python
a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 6], [7, 8]])
Ntchito yowonjezera
onjezerani = tf.onjezerani(a, b)
Ntchito zochulukitsa matrix
mul = tf.matmul(a, b)
sindikizani (f'Kuwonjezera: {kuwonjezera}')
print(f'Matrix Multiplication: {mul}')
``
5. Kupanga Chitsanzo Chosavuta cha Neural Network
Gawo lotsatira ndikupanga chitsanzo chosavuta cha netiweki ya neural. Tipanga chitsanzo chogawa zithunzi pogwiritsa ntchito deta ya MNIST, database ya zithunzi zolembedwa ndi manja. Tiyeni tiyambe:
"`python
import tensorflow ngati tf
kuchokera ku tensorflow.keras kulowetsa ma datasets, zigawo, mitundu
Kutsitsa deta ya MNIST
(zithunzi_za_sitima, zilembo_za_sitima), (zithunzi_zoyesa, zilembo_zoyesa) = ma datasets.mnist.load_data()
Kusintha kwa chithunzi
zithunzi_za_sitima, zithunzi_zoyesera = zithunzi_za_sitima / 255.0, zithunzi_zoyesera / 255.0
Kupanga chitsanzo
chitsanzo = zitsanzo. Zotsatizana([
zigawo. Zathyathyathya (zolemba_mawonekedwe=(28, 28)),
zigawo.Zothina(128, activation='relu'),
zigawo. Zokhuthala (10)
])
Kusonkhanitsa zitsanzo
model.compile(optimizer='adam',
kutayika=tf.keras.kutayika.KuchepaKuguluKudutsa mumzere(from_logits=True),
metrics=['kulondola'])
Kuphunzitsa chitsanzo
model.fit(mafano_a sitima, zilembo_za sitima, nthawi = 5)
Kuyesa chitsanzo
mayeso_otayika, mayeso_acc = chitsanzo.kuwunika (zithunzi_zoyesera, zilembo_zoyesera)
sindikizani (f'Kulondola kwa Mayeso: {test_acc}')
``
Kufotokozera:
– Ma Dataset: Timatumiza ndikuyika ma dataset a MNIST.
- Kukonzekera: Sinthani deta yanu mwa kugawa ma pixel ndi 255.
– Chitsanzo: Timatanthauzira chitsanzo chosavuta chokhala ndi zigawo ziwiri. Gawo loyamba ndi gawo la `Flatten` kuti lisinthe chithunzi cha 2D kukhala gulu la 1D. Gawo lachiwiri ndi gawo la `Dense` lomwe lili ndi ma neuron 128 ndi `relu` ngati ntchito yoyambitsa, ndipo lomaliza ndi gawo la `Dense` lomwe lili ndi ma neuron 10 omwe akuyimira magulu 10.
– Konzani: Timasonkhanitsa chitsanzo pogwiritsa ntchito `adam` optimizer ndi `SparseCategoricalCrossentropy` ngati ntchito yotayika.
– Sitima: Phunzitsani chitsanzo kwa nthawi 5.
- Yesani: Yesani chitsanzocho poyerekeza ndi deta yoyesera.
6. Ma Model Osungira ndi Kukweza
Mukamaliza kuphunzitsa chitsanzo, mungafune kuchisunga kuti mudzachigwiritse ntchito mtsogolo popanda kuchiphunzitsanso. Umu ndi momwe mungasungire ndikuyika chitsanzo:
"`python
Kusunga chitsanzo
model.save('my_model.h5')
Kutsegula chitsanzo
new_model = tf.keras.models.load_model('my_model.h5′)
Kutsimikizira chitsanzo chodzazidwa
kutayika, acc = new_model.evaluate(ma test_images, test_labels)
print(f'Kulondola kwa chitsanzo chodzaza: {acc}')
``
Mapeto
Bukuli limapereka chiyambi chatsatanetsatane choyambira ndi TensorFlow kwa oyamba kumene. Takambirana za kukhazikitsa, ntchito zoyambira za tensor, komanso kupanga chitsanzo chosavuta cha netiweki ya neural pogwiritsa ntchito deta ya MNIST. TensorFlow imapereka maluso ambiri apamwamba ofufuzira, monga kukonza deta yapamwamba, mitundu yovuta kwambiri, komanso kugwiritsa ntchito TensorFlow pazida monga TPU ndi GPU. Tikukhulupirira kuti phunziroli likuthandizani kuyamba dziko la kuphunzira makina ndi TensorFlow.
