Non hai TensorFlow sen sensores

Contido da visión xeral básicos Sobre as formas Índice Manipulación de formas Máis sobre DTypes emisións tf.convert_to para tensor Tensores xigantes Tensores de liña Tensores de aforro Tensores son arreglos multidimensionais cun tipo uniforme (chamado Podes ver todos os apoiados ao . dtype dtypes tf.dtypes Se estás familiarizado coa Os espíritos son como . Número np.arrays Todos os tensores son inmutables como números e cadeas de Python: nunca se pode actualizar o contido dun tensor, só se pode crear un novo. import tensorflow as tf import numpy as np 2024-08-15 03:05:18.327501: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:485] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2024-08-15 03:05:18.348450: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:8454] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2024-08-15 03:05:18.354825: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1452] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered básicos En primeiro lugar, crear algúns tensores básicos. Aquí está un tensor "escalar" ou "rank-0".Un escalar contén un único valor, e ningún "eixo". # This will be an int32 tensor by default; see "dtypes" below. rank_0_tensor = tf.constant(4) print(rank_0_tensor) tf.Tensor(4, shape=(), dtype=int32) WARNING: All log messages before absl::InitializeLog() is called are written to STDERR I0000 00:00:1723691120.932442 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.936343 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.940040 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.943264 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.954872 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.958376 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.961894 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.964843 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.967730 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.971300 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.974711 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691120.977717 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.208679 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.210786 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.212791 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.214776 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.216798 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.218734 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.220650 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.222554 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.224486 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.226429 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.228329 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.230251 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.269036 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.271069 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.273006 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.274956 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.276917 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.278854 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.280754 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.282664 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.284613 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.287058 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.289508 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 I0000 00:00:1723691122.291891 176945 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355 Un tensor "vector" ou "rank-1" é como unha lista de valores. # Let's make this a float tensor. rank_1_tensor = tf.constant([2.0, 3.0, 4.0]) print(rank_1_tensor) tf.Tensor([2. 3. 4.], shape=(3,), dtype=float32) A "matrix" or "rank-2" tensor has two axes: # If you want to be specific, you can set the dtype (see below) at creation time rank_2_tensor = tf.constant([[1, 2], [3, 4], [5, 6]], dtype=tf.float16) print(rank_2_tensor) tf.Tensor( [[1. 2.] [3. 4.] [5. 6.]], shape=(3, 2), dtype=float16) A scalar, shape: [] A vector, shape: [3] A matrix, shape: [3, 2] Tensors may have more axes; here is a tensor with three axes: # There can be an arbitrary number of # axes (sometimes called "dimensions") rank_3_tensor = tf.constant([ [[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]], [[10, 11, 12, 13, 14], [15, 16, 17, 18, 19]], [[20, 21, 22, 23, 24], [25, 26, 27, 28, 29]],]) print(rank_3_tensor) tf.Tensor( [[[ 0 1 2 3 4] [ 5 6 7 8 9]] [[10 11 12 13 14] [15 16 17 18 19]] [[20 21 22 23 24] [25 26 27 28 29]]], shape=(3, 2, 5), dtype=int32) Hai moitas formas de visualizar un tensor con máis de dous eixos. A 3-axis tensor, shape: [3, 2, 5] Pode converter un tensor a unha matriz NumPy usando ou o O método: np.array tensor.numpy np.array(rank_2_tensor) array([[1., 2.], [3., 4.], [5., 6.]], dtype=float16) rank_2_tensor.numpy() array([[1., 2.], [3., 4.], [5., 6.]], dtype=float16) Os sensores a miúdo conteñen flotadores e incs, pero teñen moitos outros tipos, incluíndo: Números complexos Strings A base A clase require que os tensores sexan "retangulares"---é dicir, ao longo de cada eixe, cada elemento é do mesmo tamaño. tf.Tensor Tensores Rápidos Tensores de aforro Podes facer matemáticas básicas sobre tensores, incluíndo adición, multiplicación elemental e multiplicación de matriz. a = tf.constant([[1, 2], [3, 4]]) b = tf.constant([[1, 1], [1, 1]]) # Could have also said `tf.ones([2,2], dtype=tf.int32)` print(tf.add(a, b), "\n") print(tf.multiply(a, b), "\n") print(tf.matmul(a, b), "\n") tf.Tensor( [[2 3] [4 5]], shape=(2, 2), dtype=int32) tf.Tensor( [[1 2] [3 4]], shape=(2, 2), dtype=int32) tf.Tensor( [[3 3] [7 7]], shape=(2, 2), dtype=int32) print(a + b, "\n") # element-wise addition print(a * b, "\n") # element-wise multiplication print(a @ b, "\n") # matrix multiplication tf.Tensor( [[2 3] [4 5]], shape=(2, 2), dtype=int32) tf.Tensor( [[1 2] [3 4]], shape=(2, 2), dtype=int32) tf.Tensor( [[3 3] [7 7]], shape=(2, 2), dtype=int32) Os sensores utilízanse en todo tipo de operacións (ou "Ops"). c = tf.constant([[4.0, 5.0], [10.0, 1.0]]) # Find the largest value print(tf.reduce_max(c)) # Find the index of the largest value print(tf.math.argmax(c)) # Compute the softmax print(tf.nn.softmax(c)) tf.Tensor(10.0, shape=(), dtype=float32) tf.Tensor([1 0], shape=(2,), dtype=int64) tf.Tensor( [[2.6894143e-01 7.3105854e-01] [9.9987662e-01 1.2339458e-04]], shape=(2, 2), dtype=float32) Nota: Normalmente, onde unha función TensorFlow espera un Tensor como entrada, a función tamén aceptará calquera cousa que poida ser convertida a un Tensor usando tf.convert_to_tensor. Normalmente, en calquera lugar onde unha función TensorFlow espera unha como entrada, a función tamén aceptará calquera cousa que poida converterse en Utilización Vexa abaixo para un exemplo. Note: Tensor Tensor tf.convert_to_tensor tf.convert_to_tensor([1,2,3]) tf.reduce_max([1,2,3]) tf.reduce_max(np.array([1,2,3])) Sobre as formas Algúns textos teñen vocabulario: Forma: A lonxitude (número de elementos) de cada un dos eixos dun tensor. Un escalar ten o rango 0, un vector ten o rango 1, unha matriz ten o rango 2. Eixo ou dimensión: unha dimensión particular dun tensor. Tamaño: O número total de elementos no tensor, o produto dos elementos do vector de forma. Nota: Aínda que pode ver unha referencia a un "tensor de dúas dimensións", un tensor de posición 2 normalmente non describe un espazo 2D. Aínda que se pode ver a referencia a un "tensor de dúas dimensións", un tensor de posición 2 normalmente non describe un espazo 2D. Note: Tensores e Os obxectos teñen propiedades convenientes para acceder a estes: tf.TensorShape rank_4_tensor = tf.zeros([3, 2, 4, 5]) A rank-4 tensor, shape: [3, 2, 4, 5] print("Type of every element:", rank_4_tensor.dtype) print("Number of axes:", rank_4_tensor.ndim) print("Shape of tensor:", rank_4_tensor.shape) print("Elements along axis 0 of tensor:", rank_4_tensor.shape[0]) print("Elements along the last axis of tensor:", rank_4_tensor.shape[-1]) print("Total number of elements (3*2*4*5): ", tf.size(rank_4_tensor).numpy()) Type of every element: Number of axes: 4 Shape of tensor: (3, 2, 4, 5) Elements along axis 0 of tensor: 3 Elements along the last axis of tensor: 5 Total number of elements (3*2*4*5): 120 Pero teña en conta que o e Os atributos non regresan obxectos. se precisas unha Utiliza o ou Esta diferenza é sutil, pero pode ser importante ao construír gráficos (máis tarde). Tensor.ndim Tensor.shape Tensor Tensor tf.rank tf.shape tf.rank(rank_4_tensor) tf.shape(rank_4_tensor) Moitas veces os eixos están ordenados de global a local: o eixo de lote primeiro, seguido por dimensións espaciais, e características para cada lugar por último. Typical axis order Índice Indicadores de eixo único TensorFlow segue as regras de indexación estándar de Python, similares a , e as regras básicas para a indexación NumPy. Indexar unha lista ou cadea en Python Os índices comezan en 0 os índices negativos contan cara atrás desde o final As seguintes páxinas ligan con Start:Stop:Step rank_1_tensor = tf.constant([0, 1, 1, 2, 3, 5, 8, 13, 21, 34]) print(rank_1_tensor.numpy()) [ 0 1 1 2 3 5 8 13 21 34] A indexación cun escalar elimina o eixe: print("First:", rank_1_tensor[0].numpy()) print("Second:", rank_1_tensor[1].numpy()) print("Last:", rank_1_tensor[-1].numpy()) First: 0 Second: 1 Last: 34 Índice a A liña mantén o eixe: : print("Everything:", rank_1_tensor[:].numpy()) print("Before 4:", rank_1_tensor[:4].numpy()) print("From 4 to the end:", rank_1_tensor[4:].numpy()) print("From 2, before 7:", rank_1_tensor[2:7].numpy()) print("Every other item:", rank_1_tensor[::2].numpy()) print("Reversed:", rank_1_tensor[::-1].numpy()) Everything: [ 0 1 1 2 3 5 8 13 21 34] Before 4: [0 1 1 2] From 4 to the end: [ 3 5 8 13 21 34] From 2, before 7: [1 2 3 5 8] Every other item: [ 0 1 3 8 21] Reversed: [34 21 13 8 5 3 2 1 1 0] Indicacións multiaxis Os tensores de maior rango son indexados ao pasar varios índices. As mesmas regras que no caso de un único eixe aplícanse a cada eixe de forma independente. print(rank_2_tensor.numpy()) [[1. 2.] [3. 4.] [5. 6.]] Pasando un enteiro por cada índice, o resultado é un escalar. # Pull out a single value from a 2-rank tensor print(rank_2_tensor[1, 1].numpy()) You can index using any combination of integers and slices: # Get row and column tensors print("Second row:", rank_2_tensor[1, :].numpy()) print("Second column:", rank_2_tensor[:, 1].numpy()) print("Last row:", rank_2_tensor[-1, :].numpy()) print("First item in last column:", rank_2_tensor[0, -1].numpy()) print("Skip the first row:") print(rank_2_tensor[1:, :].numpy(), "\n") Second row: [3. 4.] Second column: [2. 4. 6.] Last row: [5. 6.] First item in last column: 2.0 Skip the first row: [[3. 4.] [5. 6.]] Here is an example with a 3-axis tensor: print(rank_3_tensor[:, :, 4]) tf.Tensor( [[ 4 9] [14 19] [24 29]], shape=(3, 2), dtype=int32) Selecting the last feature across all locations in each example in the batch Ler o para aprender como aplicar a indexación para manipular elementos individuais nos seus tensores. Guía de cortes de tensión Manipulación de formas Reforzar un tensor é de gran utilidade. # Shape returns a `TensorShape` object that shows the size along each axis x = tf.constant([[1], [2], [3]]) print(x.shape) (3, 1) # You can convert this object into a Python list, too print(x.shape.as_list()) [3, 1] Pódese transformar un tecido nunha nova forma: o A operación é rápida e barata xa que os datos subxacentes non necesitan ser duplicados. tf.reshape # You can reshape a tensor to a new shape. # Note that you're passing in a list reshaped = tf.reshape(x, [1, 3]) print(x.shape) print(reshaped.shape) (3, 1) (1, 3) Os datos manteñen o seu deseño na memoria e créase un novo tensor, coa forma solicitada, apuntando aos mesmos datos.TensorFlow utiliza a orde de memoria "row-major" de estilo C, onde incrementar o índice máis dereito corresponde a un único paso na memoria. print(rank_3_tensor) tf.Tensor( [[[ 0 1 2 3 4] [ 5 6 7 8 9]] [[10 11 12 13 14] [15 16 17 18 19]] [[20 21 22 23 24] [25 26 27 28 29]]], shape=(3, 2, 5), dtype=int32) Se planeas un tensor podes ver en que orde está disposto na memoria. # A `-1` passed in the `shape` argument says "Whatever fits". print(tf.reshape(rank_3_tensor, [-1])) tf.Tensor( [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29], shape=(30,), dtype=int32) O único uso razoable da é para combinar ou dividir eixos adxacentes (ou engadir / eliminar c) A súa tf.reshape 1 Para este tensor 3x2x5, a remodelación a (3x2)x5 ou 3x(2x5) son dúas cousas razoables que facer, xa que as fendas non se mesturan: print(tf.reshape(rank_3_tensor, [3*2, 5]), "\n") print(tf.reshape(rank_3_tensor, [3, -1])) tf.Tensor( [[ 0 1 2 3 4] [ 5 6 7 8 9] [10 11 12 13 14] [15 16 17 18 19] [20 21 22 23 24] [25 26 27 28 29]], shape=(6, 5), dtype=int32) tf.Tensor( [[ 0 1 2 3 4 5 6 7 8 9] [10 11 12 13 14 15 16 17 18 19] [20 21 22 23 24 25 26 27 28 29]], shape=(3, 10), dtype=int32) Some good reshapes. A remodelación "traballará" para calquera nova forma co mesmo número total de elementos, pero non fará nada útil se non respectas a orde dos eixos. Cambio de axentes en non funciona; ten que para iso. tf.reshape tf.transpose # Bad examples: don't do this # You can't reorder axes with reshape. print(tf.reshape(rank_3_tensor, [2, 3, 5]), "\n") # This is a mess print(tf.reshape(rank_3_tensor, [5, 6]), "\n") # This doesn't work at all try: tf.reshape(rank_3_tensor, [7, -1]) except Exception as e: print(f"{type(e).__name__}: {e}") tf.Tensor( [[[ 0 1 2 3 4] [ 5 6 7 8 9] [10 11 12 13 14]] [[15 16 17 18 19] [20 21 22 23 24] [25 26 27 28 29]]], shape=(2, 3, 5), dtype=int32) tf.Tensor( [[ 0 1 2 3 4 5] [ 6 7 8 9 10 11] [12 13 14 15 16 17] [18 19 20 21 22 23] [24 25 26 27 28 29]], shape=(5, 6), dtype=int32) InvalidArgumentError: { {function_node __wrapped__Reshape_device_/job:localhost/replica:0/task:0/device:GPU:0} } Input to reshape is a tensor with 30 values, but the requested shape requires a multiple of 7 [Op:Reshape] Some bad reshapes. Pode executar sobre formas non especificadas. Ou a forma contén unha (a lonxitude dun eixe é descoñecida) ou a forma enteira é (O rango do tensor é descoñecido). None None Excepto para tf.RaggedTensor, tales formas só ocorrerán no contexto das APIs simbólicas de construción de gráficos de TensorFlow: F. Funcións O API funcional duro. Máis en DTypes Dtítulos Para comprobar a Os tipos de datos utilizan o Propiedade . tf.Tensor Tensor.dtype Ao crear unha desde un obxecto Python pode especificar opcionalmente o tipo de datos. tf.Tensor Se non, TensorFlow elixe un tipo de datos que pode representar os seus datos. e Python os números de puntos flotantes para Se non, TensorFlow usa as mesmas regras que NumPy utiliza ao converter en array. tf.int32 tf.float32 Pódese facer de tipo a tipo. the_f64_tensor = tf.constant([2.2, 3.3, 4.4], dtype=tf.float64) the_f16_tensor = tf.cast(the_f64_tensor, dtype=tf.float16) # Now, cast to an uint8 and lose the decimal precision the_u8_tensor = tf.cast(the_f16_tensor, dtype=tf.uint8) print(the_u8_tensor) tf.Tensor([2 3 4], shape=(3,), dtype=uint8) emisións A emisión é un concepto emprestado da En resumo, baixo certas condicións, os tensores máis pequenos son "estendidos" automaticamente para encaixar os tensores máis grandes cando se executan operacións combinadas neles. Funcións equivalentes en NumPy O caso máis sinxelo e máis común é cando intenta multiplicar ou engadir un tensor a un escalar. x = tf.constant([1, 2, 3]) y = tf.constant(2) z = tf.constant([2, 2, 2]) # All of these are the same computation print(tf.multiply(x, 2)) print(x * y) print(x * z) tf.Tensor([2 4 6], shape=(3,), dtype=int32) tf.Tensor([2 4 6], shape=(3,), dtype=int32) tf.Tensor([2 4 6], shape=(3,), dtype=int32) Do mesmo xeito, os eixos de lonxitude 1 poden estirarse para coincidir cos outros argumentos. Neste caso, unha matriz 3x1 é elementalmente multiplicada por unha matriz 1x4 para producir unha matriz 3x4. . [4] # These are the same computations x = tf.reshape(x,[3,1]) y = tf.range(1, 5) print(x, "\n") print(y, "\n") print(tf.multiply(x, y)) tf.Tensor( [[1] [2] [3]], shape=(3, 1), dtype=int32) tf.Tensor([1 2 3 4], shape=(4,), dtype=int32) tf.Tensor( [[ 1 2 3 4] [ 2 4 6 8] [ 3 6 9 12]], shape=(3, 4), dtype=int32) A broadcasted add: a times a gives a [3, 1] [1, 4] [3,4] Aquí está a mesma operación sen transmisión: x_stretch = tf.constant([[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3]]) y_stretch = tf.constant([[1, 2, 3, 4], [1, 2, 3, 4], [1, 2, 3, 4]]) print(x_stretch * y_stretch) # Again, operator overloading tf.Tensor( [[ 1 2 3 4] [ 2 4 6 8] [ 3 6 9 12]], shape=(3, 4), dtype=int32) A maior parte do tempo, a transmisión é eficiente tanto no tempo como no espazo, xa que a operación de transmisión nunca materializa os tensores expandidos na memoria. Vexa como funciona a emisión . tf.broadcast_to print(tf.broadcast_to(tf.constant([1, 2, 3]), [3, 3])) tf.Tensor( [[1 2 3] [1 2 3] [1 2 3]], shape=(3, 3), dtype=int32) A diferenza das matemáticas, por exemplo, Non fai nada especial para salvar a memoria. aquí, estás materializando o tensor. broadcast_to Pode ser aínda máis complicado. Descargar cancións de Jake VanderPlas mostra máis trucos de transmisión (de novo en NumPy). Esta sección Python Manual de Ciencia de Datos tf.convert_to para tensor A maioría das opcións, como e Argumentos de clase Non obstante, notarás no caso anterior que se aceptan obxectos de Python con forma de tensores. tf.matmul tf.reshape tf.Tensor A maioría, pero non todos, chamadas de ops Hai un rexistro de conversións, e a maioría das clases de obxectos como a de NumPy , Listas de Python, e Todos se converterán automaticamente. convert_to_tensor ndarray TensorShape tf.Variable que para máis detalles, e se ten o seu propio tipo que quere converter automaticamente a un tensor. tf.register_tensor_conversion_function Tensores xigantes Un tensor con números variables de elementos ao longo dalgún eixe chámase "ragged". para os datos recollidos. tf.ragged.RaggedTensor Por exemplo, Isto non pode ser representado como un tensor regular: A , shape: tf.RaggedTensor [4, None] ragged_list = [ [0, 1, 2, 3], [4, 5], [6, 7, 8], [9]] try: tensor = tf.constant(ragged_list) except Exception as e: print(f"{type(e).__name__}: {e}") ValueError: Can't convert non-rectangular Python sequence to Tensor. En vez de crear unha Utilización : tf.RaggedTensor tf.ragged.constant ragged_tensor = tf.ragged.constant(ragged_list) print(ragged_tensor) A forma dunha contén algúns eixos con lonxitudes descoñecidas: tf.RaggedTensor print(ragged_tensor.shape) (4, None) Tensores de liña é a , é dicir, pode representar datos como cadeas (arados de bytes de lonxitude variable) en tensores. tf.string dtype As cadeas son atómicas e non poden ser indexadas como as cadeas de Python. A lonxitude da cadea non é un dos eixos do tensor. para manipular as súas funcións. tf.strings Aquí está un tensor de cadea escalar: # Tensors can be strings, too here is a scalar string. scalar_string_tensor = tf.constant("Gray wolf") print(scalar_string_tensor) tf.Tensor(b'Gray wolf', shape=(), dtype=string) Un vector de cadeas: A vector of strings, shape: [3,] # If you have three string tensors of different lengths, this is OK. tensor_of_strings = tf.constant(["Gray wolf", "Quick brown fox", "Lazy dog"]) # Note that the shape is (3,). The string length is not included. print(tensor_of_strings) tf.Tensor([b'Gray wolf' b'Quick brown fox' b'Lazy dog'], shape=(3,), dtype=string) En primeiro lugar, a impresión do Prefixo indica que dtype non é unha cadea de unicode, senón unha cadea de byte. Para obter máis información sobre como traballar con texto de unicode en TensorFlow. b tf.string Unicode Tutoriais Se pasas os caracteres de unicode son UTF-8 codificados. tf.constant("🥳👍") Algunhas funcións básicas con cordas pódense atopar en incluíndo . tf.strings tf.strings.split # You can use split to split a string into a set of tensors print(tf.strings.split(scalar_string_tensor, sep=" ")) tf.Tensor([b'Gray' b'wolf'], shape=(2,), dtype=string) # ...but it turns into a `RaggedTensor` if you split up a tensor of strings, # as each string might be split into a different number of parts. print(tf.strings.split(tensor_of_strings)) Three strings split, shape: [3, None] e : tf.strings.to_number text = tf.constant("1 10 100") print(tf.strings.to_number(tf.strings.split(text, " "))) tf.Tensor([ 1. 10. 100.], shape=(3,), dtype=float32) Aínda que non podes usar Para converter un tensor de cadea en números, pode convertelo en bytes e despois en números. tf.cast byte_strings = tf.strings.bytes_split(tf.constant("Duck")) byte_ints = tf.io.decode_raw(tf.constant("Duck"), tf.uint8) print("Byte strings:", byte_strings) print("Bytes:", byte_ints) Byte strings: tf.Tensor([b'D' b'u' b'c' b'k'], shape=(4,), dtype=string) Bytes: tf.Tensor([ 68 117 99 107], shape=(4,), dtype=uint8) # Or split it up as unicode and then decode it unicode_bytes = tf.constant("アヒル 🦆") unicode_char_bytes = tf.strings.unicode_split(unicode_bytes, "UTF-8") unicode_values = tf.strings.unicode_decode(unicode_bytes, "UTF-8") print("\nUnicode bytes:", unicode_bytes) print("\nUnicode chars:", unicode_char_bytes) print("\nUnicode values:", unicode_values) Unicode bytes: tf.Tensor(b'\xe3\x82\xa2\xe3\x83\x92\xe3\x83\xab \xf0\x9f\xa6\x86', shape=(), dtype=string) Unicode chars: tf.Tensor([b'\xe3\x82\xa2' b'\xe3\x83\x92' b'\xe3\x83\xab' b' ' b'\xf0\x9f\xa6\x86'], shape=(5,), dtype=string) Unicode values: tf.Tensor([ 12450 12498 12523 32 129414], shape=(5,), dtype=int32) A súa dtype utilízase para todos os datos de bytes en bruto en TensorFlow. O módulo contén funcións para converter datos a e de bytes, incluíndo a decodificación de imaxes e a análise de csv. tf.string tf.io Tensores de aforro Ás veces, os seus datos son escasos, como un espazo de embalaxe moi amplo. e operacións relacionadas para almacenar datos escasos de forma eficiente. tf.sparse.SparseTensor A , shape: tf.SparseTensor [3, 4] # Sparse tensors store values by index in a memory-efficient manner sparse_tensor = tf.sparse.SparseTensor(indices=[[0, 0], [1, 2]], values=[1, 2], dense_shape=[3, 4]) print(sparse_tensor, "\n") # You can convert sparse tensors to dense print(tf.sparse.to_dense(sparse_tensor)) SparseTensor(indices=tf.Tensor( [[0 0] [1 2]], shape=(2, 2), dtype=int64), values=tf.Tensor([1 2], shape=(2,), dtype=int32), dense_shape=tf.Tensor([3 4], shape=(2,), dtype=int64)) tf.Tensor( [[1 0 0 0] [0 0 2 0] [0 0 0 0]], shape=(3, 4), dtype=int32) Publicado orixinalmente no sitio web de TensorFlow, este artigo aparece aquí baixo un novo título e está licenciado baixo CC BY 4.0. Publicado orixinalmente no sitio web de TensorFlow, este artigo aparece aquí baixo un novo título e está licenciado baixo CC BY 4.0. Páxina web de TensorFlow