Dzidziso yeTensorFlow kune Vanotanga
TensorFlow ndeimwe yemapuratifomu anozivikanwa zvikuru ekudzidza zvakadzama uye kudzidza kwemuchina. Yakagadzirwa nechikwata cheGoogle Brain, TensorFlow yakashandiswa zvakanyanya mumapurojekiti akawanda ekutsvagisa uye mashandisirwo emaindasitiri. Chinyorwa chino chinopa dzidziso nhanho nhanho yekukubatsira, semutangi, kutanga neTensorFlow.
1. Kunzwisisa TensorFlow Basics
Tisati tatanga kuisa nekushandisa TensorFlow, zvakakosha kuti tinzwisise kuti TensorFlow chii uye pfungwa huru dziri shure kwayo. TensorFlow ihurongwa hwakavhurika hwekuverenga nhamba uye kudzidza kwemuchina. Inoshandisa magirafu ekufambisa data kuita mashandiro enhamba, uko ma nodes ari mugirafu anomiririra mashandiro emasvomhu, uye mipendero inomiririra ma arrays edata akawanda (tensors) akabatana pakati pawo.
2. Kuiswa kweTensorFlow
Danho rekutanga mukushandisa TensorFlow kuiisa. Heino maitiro ekuisa TensorFlow uchishandisa pip, iyo Python package manager.
1. Kuiswa kwePython:
Iva nechokwadi chekuti une Python yakaiswa pasystem yako. TensorFlow inoenderana nePython 3.6 kusvika 3.9 panguva yekunyora uku. Unogona kudhawunirodha Python kubva pawebhusaiti yepamutemo yePython.
2. Nzvimbo Yepamhepo:
Zvinokurudzirwa zvikuru kugadzira nharaunda chaiyo yekuparadzanisa purojekiti yako yeTensorFlow:
“`sh
python -m venv myenv
source myenv/bin/activate Kune vashandisi veMac/Linux
myenv\Scripts\activate Kune vashandisi veWindows
``
3. Kuiswa kweTensorFlow:
Zvino, isa TensorFlow uchishandisa pombi:
“`sh
pip kuisa tensorflow
``
3. Mhoro Nyika neTensorFlow
Zvino zvayaiswa TensorFlow, ngatigadzirei Python script iri nyore kuti tione kuti yaiswa sei. Gadzirai faira idzva rePython toritumidza kuti `hello_tensorflow.py`.
"'python
import tensorflow se tf
Gadzira chinhu chisingachinji
mhoro = tf.constant('Mhoro, TensorFlow!')
Tanga chikamu
ne tf.Session() se sess:
mhedzisiro = sess.run(mhoro)
dhinda(mhedzisiro)
``
Chinja kodhi zvichienderana neTensorFlow vhezheni 2.x:
"'python
import tensorflow se tf
Gadzira chinhu chisingachinji
mhoro = tf.constant('Mhoro, TensorFlow!')
Mhanya uchishandisa eager execution (pakarepo)
printa(mhoro.numpy())
``
Sevha faira, wobva wamhanya:
“`sh
Python hello_tensorflow.py inoshandura sei Python kuti ive "hello_tensorflow"?
``
4. Kunzwisisa maTensor uye Mashandiro Ekutanga
Matensor ndiwo maumbirwo makuru edata muTensorFlow, ayo ari ma multidimensional arrays. Heano mimwe mienzaniso yekukubatsira kunzwisisa ma tensor:
"'python
import tensorflow se tf
Kugadzira ma tensor
scalar = tf. constant(7) scalar
vhekitari = tf. inogara iripo([1, 2, 3]) vhekitari
matrix = tf. nguva dzose([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) tensor ye3D
print(f'Scalar: {scalar}')
print(f'Vector: {vector}')
print(f'Matrix: {matrix}')
print(f'Tensor 3D: {tensor3d}')
``
Kuita mabasa ekutanga pa tensors:
"'python
a = tf.chinogara ([[1, 2], [3, 4]])
b = tf.chinogara ([[5, 6], [7, 8]])
Basa rekuwedzera
wedzera = tf. wedzera(a, b)
Mabasa ekuwanda kwematrix
mul = tf.matmul(a, b)
print(f'Add: {add}')
print(f'Matrix Kuwanza: {mul}')
``
5. Kugadzira Muenzaniso weNeural Network Uri Nyore
Danho rinotevera nderekugadzira modhi iri nyore yeneural network. Tichagadzira modhi yekupatsanura mifananidzo tichishandisa seti yedata reMNIST, dhatabhesi yemifananidzo yakanyorwa nemaoko. Ngatitangei:
"'python
import tensorflow se tf
kubva ku tensorflow.keras kuendesa datasets, layers, models
Kudhawunirodha seti yedata yeMNIST
(mifananidzo_yechitima, mavara_echitima), (mifananidzo_yekuedza, mavara_ekuedzwa) = datasets.mnist.load_data()
Kugadziriswa kwemufananidzo
mifananidzo_yechitima, mifananidzo_yekuyedza = mifananidzo_yechitima / 255.0, mifananidzo_yekuyedza / 255.0
Kugadzira modhi
modhi = mamodheru. Akatevedzana([
layers.Flatten(input_shape=(28, 28)),
layers.Dense(128, activation='relu'),
zvikamu. Zvakakora(10)
])
Kuunganidzwa kwemuenzaniso
model.compile(optimizer='adam',
kurasikirwa=tf.keras.kurasikirwa.KushomaKurongwaKuyambuka muchikamu(kubva_logits=Chokwadi),
metrics=['chaizvo'])
Kudzidzisa modhi
model.fit(mifananidzo_yechitima, mavara_echitima, nguva=5)
Kuedza modhi
kurasikirwa_kwekuyedza, test_acc = model.evaluate(mifananidzo_yekuyedza, mavara_ekuyedza)
print(f'Kunyatsoita bvunzo: {test_acc}')
``
Tsanangudzo:
– MaDatasets: Tinopinza uye tinoisa datasets dzeMNIST.
- Kugadzirisa: Gadzirisa dataset nekukamura kukosha kwemapikiseli ne255.
– Muenzaniso: Tinotsanangura muenzaniso uri nyore une zvikamu zviviri. Chikamu chekutanga i `Flatten` layer yekushandura mufananidzo we2D kuita 1D array. Chikamu chechipiri i `Dense` layer ine 128 neuron uye `relu` sebasa rekuita kuti zvinhu zvishande, uye chekupedzisira i `Dense` layer ine 10 neuron inomiririra makirasi gumi.
– Kuunganidza: Tinounganidza modhi tichishandisa `adam` optimizer uye `SparseCategoricalCrossentropy` sebasa rekurasikirwa.
- Chitima: Dzidzisa modhi kwemaawa mashanu.
- Ongorora: Ongorora modhi uchienzaniswa nedata rekuyedza.
6. MaModheru Ekuchengetedza Uye Kurodha
Mushure mekudzidzisa modhi, ungangoda kuichengeta kuti uzoishandisa gare gare usingazodzidzisizve. Heino maitiro ekuchengetedza nekurodha modhi:
"'python
Kuchengetedza modhi
model.save('my_model.h5')
Kurodha modhi
itsva_model = tf.keras.models.load_model('my_model.h5′)
Kusimbisa modhi yakatakura
kurasikirwa, acc = new_model.evaluate(test_images, test_labels)
print(f'Kururama kwemuenzaniso wakatakura: {acc}')
``
Mhedziso
Gwaro iri rinopa sumo yakadzama yekutanga neTensorFlow kune vanotanga. Takurukura nezvekuisa, mashandiro ekutanga etensor, uye kuvaka modhi iri nyore yeneural network tichishandisa dataset yeMNIST. TensorFlow inopa hunyanzvi hwakawanda hwepamusoro hwekuongorora, hwakadai sekugadzirisa data repamusoro, mamodheru akaomarara, uye kushandisa TensorFlow pamidziyo yakaita seTPUs neGPUs. Tinovimba kuti chidzidzo ichi chinokubatsira kutanga munyika yekudzidza kwemuchina neTensorFlow.
