Nkuzi TensorFlow maka ndị mbido
TensorFlow bụ otu n'ime usoro kachasị ewu ewu maka mmụta miri emi na mmụta igwe. Ndị otu Google Brain mepụtara TensorFlow, ejirila ya mee ihe nke ukwuu n'ọtụtụ ọrụ nyocha na ngwa ụlọ ọrụ. Edemede a na-enye nkuzi nzọụkwụ site na nzọụkwụ iji nyere gị aka, dịka onye mbido, ịmalite na TensorFlow.
1. Ịghọta Isi Ihe TensorFlow
Tupu anyị amalite itinye na iji TensorFlow, ọ dị mkpa ịghọta ihe TensorFlow bụ na echiche ndị bụ isi dị n'azụ ya. TensorFlow bụ usoro mepere emepe maka mgbakọ na mwepụ ọnụọgụgụ na mmụta igwe. Ọ na-eji eserese usoro data eme ọrụ ọnụọgụgụ, ebe nodes dị na eserese ahụ na-anọchite anya ọrụ mgbakọ na mwepụ, na nsọtụ na-anọchite anya usoro data multidimensional (tensors) ejikọtara n'etiti ha.
2. Nwụnye TensorFlow
Nzọụkwụ mbụ n'iji TensorFlow bụ itinye ya. Lee otu esi etinye TensorFlow site na iji pip, onye njikwa ngwugwu Python.
1. Nwụnye Python:
Jide n'aka na etinyere Python na sistemụ gị. TensorFlow dakọtara na Python 3.6 ruo 3.9 n'oge a na-ede ihe a. Ị nwere ike ibudata Python na weebụsaịtị Python gọọmentị.
2. Gburugburu Ebe Obibi Mebere Emebere:
A na-atụ aro nke ukwuu ka ịmepụta gburugburu ebe obibi mebere iji kewapụ ọrụ TensorFlow gị:
"`sh"
Python - m venv myenv
isi mmalite myenv/bin/activate Maka ndị ọrụ Mac/Linux
myenv\Scripts\activate Maka ndị ọrụ Windows
““
3. Ntinye TensorFlow:
Ugbu a, wụnye TensorFlow site na iji pip:
"`sh"
pip tinye tensorflow
““
3. Ndewo Ụwa na TensorFlow
Ugbua etinyere TensorFlow, ka anyị mepụta edemede Python dị mfe iji chọpụta nrụnye ahụ. Mepụta faịlụ Python ọhụrụ wee kpọọ ya `hello_tensorflow.py`.
"'Python
mbubata tensorflow dị ka tf
Mepụta ihe na-agbanwe agbanwe
ndewo = tf.constant('Ndewo, TensorFlow!')
Malite nnọkọ
na tf.Session() dị ka sess:
nsonaazụ = sess.gba ọsọ(ndewo)
ebipụta(nsonaazụ)
““
Gbanwee koodu ahụ dịka ụdị TensorFlow 2.x si dị:
"'Python
mbubata tensorflow dị ka tf
Mepụta ihe na-agbanwe agbanwe
ndewo = tf.constant('Ndewo, TensorFlow!')
Gbaa ọsọ site na iji ihe eji eme ihe nke ọma (gbanye ya na ndabara)
bipụta(ndeewo.numpy())
““
Chekwaa faịlụ ahụ, wee mee ọsọ:
"`sh"
Python hello_tensorflow.py
““
4. Ịghọta Tensors na Ọrụ Ndị Dị Mkpa
Tensọ bụ isi nhazi data dị na TensorFlow, nke bụ ọtụtụ nhazi. Lee ụfọdụ ihe atụ iji nyere gị aka ịghọta tensọ:
"'Python
mbubata tensorflow dị ka tf
Ịmepụta tensors
scalar = tf. scalar na-agbanweghi agbanwe(7)
vektọ = tf. na-agbanwe agbanwe([1, 2, 3]) vektọ
matriks = tf. constant([[1, 2], [3, 4]]) matriks
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) tensọ 3D
bipụta (f'Scalar: {scalar}')
bipụta (f'Vector: {vector}')
bipụta(f'Matrix: {matrix}')
bipụta(f'Tensor 3D: {tensor3d}')
““
Iji rụọ ọrụ ndị bụ isi na tensors:
"'Python
a = tf.na-agbanwe agbanwe([[1, 2], [3, 4]])
b = tf.na-agbanwe agbanwe([[5, 6], [7, 8]])
Ọrụ mgbakwunye
tinye = tf.tinye(a, b)
Ọrụ mmụba matriks
mul = tf.matmul(a, b)
bipụta(f'Mgbakwunye: {tinye}')
bipụta(f'Mmụba Matrix: {mul}')
““
5. Ịmepụta Ụdị Netwọk Neural Dị Mfe
Nzọụkwụ ọzọ bụ ịmepụta ụdị netwọkụ akwara dị mfe. Anyị ga-ewulite ụdị nhazi onyonyo site na iji dataset MNIST, nchekwa data nke onyonyo ọnụọgụgụ e dere aka. Ka anyị malite:
"'Python
mbubata tensorflow dị ka tf
site na tensorflow.keras mbubata data, oyi akwa, na ụdị
Ibudata data MNIST
(foto_ụgbọ okporo ígwè, akara_ụgbọ okporo ígwè), (foto_ụgbọ okporo ígwè, akara_ụgbọ ule) = datasets.mnist.load_data()
Nhazi onyonyo
onyonyo_ụgbọ okporo ígwè, onyonyo_ule = onyonyo_ụgbọ okporo ígwè / 255.0, onyonyo_ule / 255.0
Ime ihe nlereanya
ụdị = ụdị. Usoro([
oyi akwa. Gbasaa (ọdịdị_ntinye=(28, 28)),
layers.Dense(128, activation='relu'),
oyi akwa.Ogwu(10)
])
Nchịkọta ihe nlereanya
model.compile(njikarịcha ='adam',
mfu=tf.keras.mfu.SparseCategoricalCrosentropy(site na_logits=Eziokwu),
metrics=['nke ziri ezi'])
Ịzụ ihe nlereanya ahụ
model.fit (ihe oyiyi ụgbọ oloko, akara ụgbọ oloko, epochs=5)
Na-anwale ihe nlereanya ahụ
nnwale_mfu, ule_acc = ụdị.nyocha(onyonyo_ule, akara_ule)
bipụta(f'Izi ezi ule: {test_acc}')
““
Nkọwa:
– Datasets: Anyị na-ebubata ma na-ebu dataset MNIST.
- Nhazi tupu oge eruo: Mee ka dataset dị mma site na ikewa uru pikselụ site na 255.
– Ụdị: Anyị na-akọwa ụdị dị mfe nke nwere akwa abụọ. Nke mbụ bụ akwa 'Flatten' iji gbanwee onyonyo 2D ka ọ bụrụ usoro 1D. Nke abụọ bụ akwa 'Dense' nwere neuron 128 na 'relu' dị ka ọrụ mmalite, nke ikpeazụ bụ akwa 'Dense' nwere neuron 10 na-anọchite anya klas 10.
– Chịkọta: Anyị na-eji ihe na-eme ka ihe nlereanya ahụ dị mma site na iji ihe na-eme ka ihe dịkwuo mma (adam` optimizer) na ihe na-eme ka ihe dị iche iche dị ka "SparseCategoricalCrossentropy" dị ka ọrụ mfu.
– Ụgbọ oloko: Zụọ ihe nlereanya ahụ maka oge ise.
– Nyochaa: Nyochaa ihe nlereanya ahụ site na data nnwale.
6. Ịchekwa na Ibugo Ụdị
Mgbe ị zụrụ ụdị nlereanya, ị nwere ike ịchọrọ ịchekwa ya maka ojiji ma emechaa n'emeghị ka ọ bụrụ ọzụzụ ọzọ. Lee otu esi echekwa ma tinye ụdị nlereanya:
"'Python
Ịchekwa ụdị ihe nlereanya ahụ
ụdị.chekwaa('m_model.h5')
Ụdị na-ebugo
ụdị_ọhụrụ = tf.keras.ụdị.ụdị_ibu('m_ụdị_m5′)
Ịchọpụta ụdị ebugoro ibu
mfu, acc = ụdị_ọhụrụ.evaluate( onyonyo_ule, akara_ule)
bipụta(f'Izi ezi nke ụdị ebugoro: {acc}')
““
Mmechi
Nduzi a na-enye nkọwa zuru ezu gbasara ịmalite TensorFlow maka ndị mbido. Anyị ekpuchila ntinye, ọrụ tensor bụ isi, na iwulite ụdị netwọkụ akwara dị mfe site na iji dataset MNIST. TensorFlow na-enye ọtụtụ ikike dị elu inyocha, dị ka nhazi data dị elu, ụdị dị mgbagwoju anya karị, na iji TensorFlow na ngwaọrụ dị ka TPUs na GPUs. Anyị nwere olileanya na nkuzi a ga-enyere gị aka ịmalite n'ụwa mmụta igwe na TensorFlow.
