Thuto ea TensorFlow bakeng sa ba qalang
TensorFlow ke e 'ngoe ea meralo e tsebahalang haholo bakeng sa ho ithuta ka botebo le ho ithuta ka mochini. E ntlafalitsoe ke sehlopha sa Google Brain, TensorFlow e sebelisitsoe haholo mererong e mengata ea lipatlisiso le lits'ebetsong tsa indasteri. Sengoloa sena se fana ka thupelo ea mohato ka mohato ho u thusa, joalo ka moqali, ho qala ka TensorFlow.
1. Ho Utloisisa Metheo ea TensorFlow
Pele re qala ho kenya le ho sebedisa TensorFlow, ho bohlokwa ho utlwisisa hore na TensorFlow ke eng le mehopolo ya motheo e ka morao ho yona. TensorFlow ke moralo wa mohlodi o bulehileng bakeng sa ho bala dipalo le ho ithuta ha mochini. E sebedisa dikerafo tsa phallo ya data ho etsa mesebetsi ya dipalo, moo dinode tse kerafong di emelang mesebetsi ya dipalo, mme mathoko a emela di-array tsa data tse nang le mahlakore a mangata (di-tensor) tse hokahantsweng pakeng tsa tsona.
2. Ho kenya TensorFlow
Mohato oa pele oa ho sebelisa TensorFlow ke ho e kenya. Mona ke mokhoa oa ho kenya TensorFlow u sebelisa pip, molaoli oa sephutheloana sa Python.
1. Ho kenya Python:
Etsa bonnete ba hore o kentse Python tsamaisong ea hau. TensorFlow e tsamaisana le Python 3.6 ho isa ho 3.9 nakong ea ho ngola sena. U ka khoasolla Python ho tsoa webosaeteng ea semmuso ea Python.
2. Tikoloho ea Inthanete:
Ho kgothaletswa haholo ho theha tikoloho ya sebele ho arola projeke ya hao ya TensorFlow:
“`sh
python -m venv myenv
mohloli myenv/bin/activate Bakeng sa basebelisi ba Mac/Linux
myenv\Scripts\activate Bakeng sa basebelisi ba Windows
``
3. Ho kenya TensorFlow:
Joale, kenya TensorFlow u sebelisa pip:
“`sh
kenya tensorflow ea pip
``
3. Lumela Lefatše ka TensorFlow
Jwale kaha TensorFlow e kentswe, ha re etseng sengolwa se bonolo sa Python ho netefatsa ho kenngwa. Theha faele e ntjha ya Python mme o e rehe lebitso la `hello_tensorflow.py`.
"'python
import tensorflow joalo ka tf
Theha ntho e sa fetoheng
dumela = tf.constant('Lumela, TensorFlow!')
Qala thuto
ka tf.Session() jwalo ka sess:
sephetho = sess.run (lumelisa)
hatisa(sephetho)
``
Fetola khoutu ho latela mofuta oa TensorFlow 2.x:
"'python
import tensorflow joalo ka tf
Theha ntho e sa fetoheng
dumela = tf.constant('Lumela, TensorFlow!')
Matha o sebelisa ts'ebetso e labalabelang (ka boiketsetso)
printa(hello.numpy())
``
Boloka faele, ebe o matha:
“`sh
python hello_tensorflow.py
``
4. Ho Utloisisa Li-Tensor le Ts'ebetso ea Motheo
Li-tensor ke sebopeho sa mantlha sa data ho TensorFlow, e leng li-array tse nang le mahlakore a mangata. Mehlala e meng ke ena ho u thusa ho utloisisa li-tensor:
"'python
import tensorflow joalo ka tf
Ho theha li-tensor
scalar = tf. kamehla(7) scalar
vekthara = tf. kamehla([1, 2, 3]) vekthara
matrix = tf. kamehla([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) tensor ea 3D
print(f'Scalar: {scalar}')
print(f'Vektha: {vektha}')
print(f'Matrix: {matrix}')
print(f'Tensor 3D: {tensor3d}')
``
Ho etsa mesebetsi ea motheo holim'a li-tensor:
"'python
a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 6], [7, 8]])
Ts'ebetso ea ho eketsa
eketsa = tf. eketsa(a, b)
Mesebetsi ea ho atisa matrix
mul = tf.matmul(a, b)
printa(f'Add: {add}')
print(f'Matrix Atisa: {mul}')
``
5. Ho theha Mohlala o Bonolo oa Marangrang a Methapo
Mohato o latelang ke ho theha mohlala o bonolo oa marangrang a methapo. Re tla haha mohlala oa ho arola litšoantšo re sebelisa sete ea data ea MNIST, database ea litšoantšo tse ngotsoeng ka letsoho tsa linomoro. A re qaleng:
"'python
import tensorflow joalo ka tf
ho tsoa ho tensorflow.keras kenya li-dataset, mekhahlelo, mehlala
Ho khoasolla sete ea data ea MNIST
(litšoantšo_tsa_terene, mabitso_a_terene), (litšoantšo_tsa_teko, mabitso_a_teko) = li-dataset.mnist.load_data()
Ho tloaeleha ha setšoantšo
litšoantšo_tsa_terene, litšoantšo_tsa_teko = litšoantšo_tsa_terene / 255.0, litšoantšo_tsa_teko / 255.0
Ho etsa mohlala
mohlala = mehlala.E latellanang([
mekhahlelo. Batalatsa (sebopeho_sa_input=(28, 28)),
layers.E teteaneng(128, activation='relu'),
mekhahlelo. E teteaneng (10)
])
Pokello ea mehlala
model.compile(optimizer='adam',
tahlehelo=tf.keras.tahlehelo.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
Ho koetlisa mohlala
model.fit(litšoantšo_tsa_terene, mabitso_a_terene, linako=5)
Ho leka mohlala
tahlehelo_ya_teko, teko_acc = mohlala.evaluate(litšoantšo_tsa_teko, mabitso_a_teko)
print(f'Ho nepahala ha teko: {test_acc}')
``
Tlhaloso:
– Disete tsa Dintlha: Re kenya le ho kenya sete ya data ya MNIST.
– Ho sebetsana le data esale pele: Fetola boleng ba data ka ho arola boleng ba pixel ka 255.
– Mohlala: Re hlalosa mohlala o bonolo o nang le mekhahlelo e 'meli. Lera la pele ke lera la `Flatten` ho fetolela setšoantšo sa 2D hore e be sehlopha sa 1D. Lera la bobeli ke lera la `Dense` le nang le li-neurone tse 128 le `relu` e le mosebetsi oa ho kenya tšebetsong, 'me la ho qetela ke lera la `Dense` le nang le li-neurone tse 10 tse emelang lihlopha tse 10.
– Ho bokella: Re bokella mohlala re sebedisa `adam` optimizer le `SparseCategoricalCrossentropy` e le mosebetsi wa tahlehelo.
– Terene: Koetlisa mohlala ka linako tse 5.
– Lekola: Lekola mohlala khahlanong le lintlha tsa teko.
6. Mehlala ea ho Boloka le ho Jarisa
Kamora ho koetlisa mohlala, o kanna oa batla ho o boloka bakeng sa tšebeliso ea morao ntle le ho o koetlisa hape. Mokhoa oa ho boloka le ho kenya mohlala ke ona:
"'python
Ho boloka mohlala
mohlala.boloka('mohlala_wa_ka ...
Ho kenya mohlala
mohlala_o_motjha = tf.keras.models.load_model('mohlala_wa_ka_ka.h5′)
Ho netefatsa mohlala o jarollotsoeng
tahlehelo, acc = new_model.evaluate(litšoantšo_tsa_teko, mabitso_a_teko)
print(f'Ho nepahala ha mohlala o jarollotsweng: {acc}')
``
Qetello
Tataiso ena e fana ka selelekela se qaqileng sa ho qala ka TensorFlow bakeng sa ba qalang. Re se re akarelitse ho kenya, ts'ebetso ea motheo ea tensor, le ho aha mohlala o bonolo oa marang-rang a methapo re sebelisa sete ea data ea MNIST. TensorFlow e fana ka bokhoni bo bongata bo tsoetseng pele ba ho bo hlahloba, joalo ka ts'ebetso e tsoetseng pele ea data, mehlala e rarahaneng haholoanyane, le ho sebelisa TensorFlow lisebelisoa tse kang TPU le GPU. Re tšepa hore thupelo ena e u thusa ho qala lefats'eng la ho ithuta ka mochini ka TensorFlow.
