Casharka TensorFlow ee loogu talagalay Bilawga
TensorFlow waa mid ka mid ah qaababka ugu caansan ee waxbarashada qoto dheer iyo barashada mashiinka. Waxaa sameeyay kooxda Google Brain, TensorFlow waxaa si weyn loogu isticmaalay mashaariic cilmi baaris oo badan iyo codsiyada warshadaha. Maqaalkani wuxuu bixiyaa casharro tallaabo-tallaabo ah si uu kaaga caawiyo, bilow ahaan, inaad ku bilowdo TensorFlow.
1. Fahmidda Aasaaska TensorFlow
Kahor inta aynaan bilaabin rakibidda iyo isticmaalka TensorFlow, waa muhiim in la fahmo waxa TensorFlow yahay iyo fikradaha aasaasiga ah ee ka dambeeya. TensorFlow waa qaab-dhismeed il furan oo loogu talagalay xisaabinta tirooyinka iyo barashada mashiinka. Waxay isticmaashaa garaafyada socodka xogta si ay u qabato hawlgallada tirooyinka, halkaas oo qanjidhada jaantusku ay matalaan hawlgallada xisaabta, geesahana ay matalaan taxane xog badan (tensors) oo isku xiran.
2. Rakibaadda TensorFlow
Tallaabada ugu horreysa ee isticmaalka TensorFlow waa rakibiddeeda. Waa kan sida loo rakibo TensorFlow iyadoo la adeegsanayo pip, oo ah maamulaha baakadka Python.
1. Rakibaadda Python:
Hubi inaad Python ku rakibtay nidaamkaaga. TensorFlow wuxuu la jaan qaadayaa Python 3.6 ilaa 3.9 waqtiga qoraalkan. Waxaad Python ka soo dejisan kartaa bogga rasmiga ah ee Python.
2. Deegaanka Dalwaddeed:
Waxaa aad loogu talinayaa inaad abuurto jawi dalwaddeed si aad u go'doomiso mashruucaaga TensorFlow:
"`sh"
Python -m venv myenv
Isha myenv/bin/activate Loogu talagalay dadka isticmaala Mac/Linux
myenv\Scripts\activate Loogu talagalay dadka isticmaala Windows
““
3. Rakibaadda TensorFlow:
Hadda, ku rakib TensorFlow adoo isticmaalaya pip:
"`sh"
pip install tensorflow
““
3. Salaan Adduunka oo leh TensorFlow
Hadda oo TensorFlow la rakibay, aan samayno qoraal Python fudud si aan u xaqiijino rakibidda. Samee fayl Python cusub oo u bixi `hello_tensorflow.py`.
"'python
soo dejinta tensorflow sida tf
Samee joogto ah
hello = tf.constant('Hello, TensorFlow!')
Bilow kalfadhiga
iyadoo la adeegsanayo tf.Session() sida sess:
natiijo = sess.orod(hello)
daabac(natiijo)
““
Ku habboonow koodhka sida waafaqsan nooca TensorFlow ee 2.x:
"'python
soo dejinta tensorflow sida tf
Samee joogto ah
hello = tf.constant('Hello, TensorFlow!')
Ku orod adigoo adeegsanaya fulinta xiisaha leh (sida caadiga ah ku shid)
daabac(hello.numpy())
““
Kaydi faylka, ka dibna socodsii:
"`sh"
Python hello_tensorflow.py
““
4. Fahmidda Tensor-yada iyo Hawlgallada Aasaasiga ah
Tensors-ku waa qaab-dhismeedka xogta ugu muhiimsan ee TensorFlow, kuwaas oo ah tiro badan oo kala-duwan. Waa kuwan tusaalooyin kaa caawinaya inaad fahamto tensor-yada:
"'python
soo dejinta tensorflow sida tf
Abuuritaanka tensors
scalar = tf. joogto ah (7) scalar
vektor = tf. joogto ah([1, 2, 3]) vektor
matrix = tf. joogto ah([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) Tensor 3D ah
daabac(f'Scalar: {scalar}')
daabac(f'Vector: {vector}')
daabac(f'Matrix: {matrix}')
daabac(f'Tensor 3D: {tensor3d}')
““
Si loo sameeyo hawlgallo aasaasi ah oo ku saabsan tensor-yada:
"'python
a = tf.joogto ah([[1, 2], [3, 4]])
b = tf.joogto ah([[5, 6], [7, 8]])
Hawlgalka dheeraadka ah
ku dar = tf. ku dar(a, b)
Hawlgallada isku-dhufashada Matrix-ka
mul = tf.matmul(a, b)
daabac(f'Ku darid: {ku dar}')
daabac(f'Isku-dhufashada Matrix: {mul}')
““
5. Abuuritaanka Qaab Shabakad Dareeme oo Fudud
Tallaabada xigta waa in la abuuro qaab shabakadeed oo fudud. Waxaan dhisi doonnaa qaab kala soocid sawireed annagoo adeegsanayna xogta MNIST, oo ah xog ururin sawirro gacmeed lagu qoray. Aan bilowno:
"'python
soo dejinta tensorflow sida tf
laga bilaabo tensorflow.keras soo dejinta xog ururinta, lakabyada, moodooyinka
Soo dejinta xogta MNIST
(sawirrada_tareenka, calaamadaha_tareenka), (sawirrada_tijaabada, calaamadaha_tijaabada) = xogta xogta.mnist.xogta_rarka()
Habaynta sawirka
sawirada_tareenka, sawirada_tijaabada = sawirada_tareenka / 255.0, sawirada_tijaabada / 255.0
Samaynta qaab
model = models.Taxane([
Lakabyo. Kala fidi(qaabka_gelinta=(28, 28)),
lakabyo.Dense(128, activation='relu'),
Lakabyo.Cufan (10)
])
Isku-dubaridka moodalka
model.compile(optimizer='adam',
khasaare=tf.keras.khasaaro.SparseQaybtaIsku-xidhkaCrossentropy(laga bilaabo_logits=Run),
metrics=['sax'])
Tababaridda qaabka
model.ku habboon (sawirrada_tareenka, calaamadaha_tareenka, xilliyada=5)
Tijaabinta qaabka
tijaabo_khasaare, tijaabo_acc = moodel.qiimee(sawirrada_tijaabada, calaamadaha_tijaabada)
daabac(f'Saxnaanta Tijaabada: {test_acc}')
““
Sharaxaad:
– Xogta: Waxaan soo dejinnaa oo rarnaa xogta MNIST.
- Horusocod: Caadiyeeya xogta xogta iyadoo loo qaybinayo qiimaha pixel-ka 255.
– Qaabka: Waxaan qeexnaa qaab fudud oo leh laba lakab. Lakabka koowaad waa lakabka 'Flatten' si loogu beddelo sawirka 2D taxane 1D ah. Lakabka labaad waa lakabka 'Censive' oo leh 128 neerfood iyo 'relu' oo ah shaqada kicinta, kan ugu dambeeyana waa lakabka 'Censive' oo leh 10 neerfood oo matalaya 10 fasal.
– Soo ururi: Waxaan qaab-dhismeedka ku soo ururinaa annagoo adeegsanayna 'adam' optimizer iyo 'SparseCategoricalCrossentropy' oo ah shaqada khasaaraha.
– Tareen: Tababar qaabka 5 jeer.
– Qiimee: Qiimee qaabka iyadoo la eegayo xogta tijaabada.
6. Noocyada Kaydinta iyo Soo-rarista
Ka dib markaad tababarto moodel, waxaa laga yaabaa inaad rabto inaad kaydsato si aad u isticmaasho dambe adigoon dib u tababarin. Waa kan sida loo keydiyo loona raro moodel:
"'python
Keydinta qaabka
model.keydi('my_model.h5')
Qaabka la soo dejinayo
qaab_cusub = tf.keras.model.load_model('my_model.h5′)
Xaqiijinta qaabka la raray
khasaare, acc = qaab_cusub.qiimayn (sawirrada_tijaabada, calaamadaha_tijaabada)
daabacan(f'Saxnaanta qaabka la raray: {acc}')
““
Gabagabo
Tilmaamahan wuxuu bixinayaa hordhac faahfaahsan oo ku saabsan bilaabista TensorFlow ee loogu talagalay kuwa bilowga ah. Waxaan daboolnay rakibidda, hawlgallada tensor-ka aasaasiga ah, iyo dhisidda qaab shabakad neerfeed oo fudud iyadoo la adeegsanayo xogta MNIST. TensorFlow waxay bixisaa awoodo badan oo horumarsan oo lagu sahaminayo, sida habaynta xogta horumarsan, moodooyinka aadka u adag, iyo isticmaalka TensorFlow ee aaladaha sida TPUs iyo GPUs. Waxaan rajeyneynaa in casharkan uu kaa caawin doono inaad ku bilowdo adduunka barashada mashiinka TensorFlow.
