Tensor Slicing and Data Insertion Made Easy with TensorFlow (TensorFlowで簡単にテナンスとデータ入力)

コンテンツ概要 導入 TENSOR スライス TENSOR にデータを挿入する さらなる読書とリソース オブジェクト検出やNLPなどのMLアプリケーションで作業する場合、たとえば、あなたのモデルアーキテクチャにルーティングが含まれている場合、1つのレイヤーが、どのトレーニングサンプルが次のレイヤーにルーティングされるかを制御する場合があります。 NLP アプリケーションでは、トレーニング中にテンサー切断を使用して単語マッキングを実行できます. For example, you can generate training data from a list of sentences by choosing a word index to mask in each sentence, taking the word out as a label, and then replacing the chosen word with a mask token. 例えば、あなたは、単語のトレーニングデータを単語のリストから選択して単語のインデックスを選択して、各句をマスクし、単語をラベルとして取り出し、その後、選択した単語をマスクトークンに置き換えることができます。 このガイドでは、TensorFlow API を使用する方法を学びます: テナーからシリーズを抽出 Tensor 内の特定のインデックスにデータを挿入する このガイドは、テナント指数化に精通していると仮定します。 そして このガイドを始める前に テンサー TensorFlow ナンパイ セットアップ import tensorflow as tf import numpy as np 2024-08-15 02:59:22.387959: 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 02:59:22.409014: 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 02:59:22.415391: 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 TENSOR スライス NumPy-like tensor slicing を使用する . tf.slice t1 = tf.constant([0, 1, 2, 3, 4, 5, 6, 7]) print(tf.slice(t1, begin=[1], size=[3])) tf.Tensor([1 2 3], shape=(3,), dtype=int32) WARNING: All log messages before absl::InitializeLog() is called are written to STDERR I0000 00:00:1723690765.029742 169903 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 00:00:1723690765.033507 169903 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 00:00:1723690765.037154 169903 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 00:00:1723690765.040869 169903 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 00:00:1723690765.052765 169903 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 00:00:1723690765.056359 169903 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 00:00:1723690765.059719 169903 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 00:00:1723690765.063138 169903 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 00:00:1723690765.066532 169903 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 00:00:1723690765.070040 169903 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 00:00:1723690765.073650 169903 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 00:00:1723690765.077235 169903 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 00:00:1723690766.300773 169903 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 00:00:1723690766.302797 169903 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 00:00:1723690766.304766 169903 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 00:00:1723690766.306851 169903 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 00:00:1723690766.308843 169903 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 00:00:1723690766.310723 169903 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 00:00:1723690766.312589 169903 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 00:00:1723690766.314574 169903 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 00:00:1723690766.316423 169903 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 00:00:1723690766.318291 169903 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 00:00:1723690766.320168 169903 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 00:00:1723690766.322150 169903 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 00:00:1723690766.360968 169903 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 00:00:1723690766.362946 169903 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 00:00:1723690766.364849 169903 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 00:00:1723690766.366867 169903 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 00:00:1723690766.368710 169903 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 00:00:1723690766.370591 169903 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 00:00:1723690766.372447 169903 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 00:00:1723690766.374435 169903 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 00:00:1723690766.376302 169903 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 00:00:1723690766.378677 169903 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 00:00:1723690766.380966 169903 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 00:00:1723690766.383365 169903 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 Alternatively, you can use a more Pythonic syntax. Note that tensor slices are evenly spaced over a start-stop range. 代わりに、よりピトニックなシンタクスを使用できます。 print(t1[1:4]) tf.Tensor([1 2 3], shape=(3,), dtype=int32) print(t1[-3:]) tf.Tensor([5 6 7], shape=(3,), dtype=int32) 二次元のテネスには、以下のようなものを用いることができます。 t2 = tf.constant([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9], [10, 11, 12, 13, 14], [15, 16, 17, 18, 19]]) print(t2[:-1, 1:3]) tf.Tensor( [[ 1 2] [ 6 7] [11 12]], shape=(3, 2), dtype=int32) 利用できます♪ さらに高次元のテントも。 tf.slice t3 = tf.constant([[[1, 3, 5, 7], [9, 11, 13, 15]], [[17, 19, 21, 23], [25, 27, 29, 31]] ]) print(tf.slice(t3, begin=[1, 1, 0], size=[1, 1, 2])) tf.Tensor([[[25 27]]], shape=(1, 1, 2), dtype=int32) You can also use テネザーのサイズを「ストライディング」することで、テネザーの切片を抽出する。 tf.strided_slice 利用 Tensor の単一の軸から特定のインデックスを抽出する。 tf.gather print(tf.gather(t1, indices=[0, 3, 6])) # This is similar to doing t1[::3] tf.Tensor([0 3 6], shape=(3,), dtype=int32) 指数を均等に分割する必要はありません。 tf.gather alphabet = tf.constant(list('abcdefghijklmnopqrstuvwxyz')) print(tf.gather(alphabet, indices=[2, 0, 19, 18])) tf.Tensor([b'c' b'a' b't' b's'], shape=(4,), dtype=string) Tensor の複数の軸からスライスを抽出するには、 これは、単にその行または列の代わりに、マトリックスの要素を集めたいときに有用です。 tf.gather_nd t4 = tf.constant([[0, 5], [1, 6], [2, 7], [3, 8], [4, 9]]) print(tf.gather_nd(t4, indices=[[2], [3], [0]])) tf.Tensor( [[2 7] [3 8] [0 5]], shape=(3, 2), dtype=int32) t5 = np.reshape(np.arange(18), [2, 3, 3]) print(tf.gather_nd(t5, indices=[[0, 0, 0], [1, 2, 1]])) tf.Tensor([ 0 16], shape=(2,), dtype=int64) # Return a list of two matrices print(tf.gather_nd(t5, indices=[[[0, 0], [0, 2]], [[1, 0], [1, 2]]])) tf.Tensor( [[[ 0 1 2] [ 6 7 8]] [[ 9 10 11] [15 16 17]]], shape=(2, 2, 3), dtype=int64) # Return one matrix print(tf.gather_nd(t5, indices=[[0, 0], [0, 2], [1, 0], [1, 2]])) tf.Tensor( [[ 0 1 2] [ 6 7 8] [ 9 10 11] [15 16 17]], shape=(4, 3), dtype=int64) TENSOR にデータを挿入する 利用 値を挿入するテンサーの特定のシリーズ/インデックスにデータを挿入するには、値を挿入するテンサーがゼロに初期化されていることに注意してください。 tf.scatter_nd t6 = tf.constant([10]) indices = tf.constant([[1], [3], [5], [7], [9]]) data = tf.constant([2, 4, 6, 8, 10]) print(tf.scatter_nd(indices=indices, updates=data, shape=t6)) tf.Tensor([ 0 2 0 4 0 6 0 8 0 10], shape=(10,), dtype=int32) 方法像 ゼロ初期化のテンサーを必要とするものは、希少なテンサー初期化器に似ています。 そして sparse tensor opsの動作を模する。 tf.scatter_nd tf.gather_nd tf.scatter_nd これらの2つの方法を組み合わせたスパーステナーを構築する例を考えてみましょう。 # Gather values from one tensor by specifying indices new_indices = tf.constant([[0, 2], [2, 1], [3, 3]]) t7 = tf.gather_nd(t2, indices=new_indices) # Add these values into a new tensor t8 = tf.scatter_nd(indices=new_indices, updates=t7, shape=tf.constant([4, 5])) print(t8) tf.Tensor( [[ 0 0 2 0 0] [ 0 0 0 0 0] [ 0 11 0 0 0] [ 0 0 0 18 0]], shape=(4, 5), dtype=int32) これは似ている: t9 = tf.SparseTensor(indices=[[0, 2], [2, 1], [3, 3]], values=[2, 11, 18], dense_shape=[4, 5]) print(t9) SparseTensor(indices=tf.Tensor( [[0 2] [2 1] [3 3]], shape=(3, 2), dtype=int64), values=tf.Tensor([ 2 11 18], shape=(3,), dtype=int32), dense_shape=tf.Tensor([4 5], shape=(2,), dtype=int64)) # Convert the sparse tensor into a dense tensor t10 = tf.sparse.to_dense(t9) print(t10) tf.Tensor( [[ 0 0 2 0 0] [ 0 0 0 0 0] [ 0 11 0 0 0] [ 0 0 0 18 0]], shape=(4, 5), dtype=int32) 既存の値を持つテンサーにデータを挿入するには、 . tf.tensor_scatter_nd_add t11 = tf.constant([[2, 7, 0], [9, 0, 1], [0, 3, 8]]) # Convert the tensor into a magic square by inserting numbers at appropriate indices t12 = tf.tensor_scatter_nd_add(t11, indices=[[0, 2], [1, 1], [2, 0]], updates=[6, 5, 4]) print(t12) tf.Tensor( [[2 7 6] [9 5 1] [4 3 8]], shape=(3, 3), dtype=int32) 同様に、使用 既存の値を含むテンサーから値を引き下げる。 tf.tensor_scatter_nd_sub # Convert the tensor into an identity matrix t13 = tf.tensor_scatter_nd_sub(t11, indices=[[0, 0], [0, 1], [1, 0], [1, 1], [1, 2], [2, 1], [2, 2]], updates=[1, 7, 9, -1, 1, 3, 7]) print(t13) tf.Tensor( [[1 0 0] [0 1 0] [0 0 1]], shape=(3, 3), dtype=int32) 利用 元素型の最小値を一つのテナントから別のテナントにコピーする。 tf.tensor_scatter_nd_min t14 = tf.constant([[-2, -7, 0], [-9, 0, 1], [0, -3, -8]]) t15 = tf.tensor_scatter_nd_min(t14, indices=[[0, 2], [1, 1], [2, 0]], updates=[-6, -5, -4]) print(t15) tf.Tensor( [[-2 -7 -6] [-9 -5 1] [-4 -3 -8]], shape=(3, 3), dtype=int32) 同様に、使用 元素型の最大値を一つのテンサーからもう一つにコピーする。 tf.tensor_scatter_nd_max t16 = tf.tensor_scatter_nd_max(t14, indices=[[0, 2], [1, 1], [2, 0]], updates=[6, 5, 4]) print(t16) tf.Tensor( [[-2 -7 6] [-9 5 1] [ 4 -3 -8]], shape=(3, 3), dtype=int32) さらなる読書とリソース このガイドでは、TensorFlowで利用可能なTensorSlicingオプションを使用して、TensorFlowの要素をより優れた制御する方法を学びました。 tf.experimental.numpy.take_along_axis および tf.experimental.numpy.take などの TensorFlow NumPy で利用可能な削除オプションをご覧ください。 また、TensorガイドとVariableガイドをご覧ください。 最初はTensorFlowのウェブサイトに掲載され、この記事はここに新しいタイトルの下に表示され、CC BY 4.0でライセンスを受けています。 最初はTensorFlowのウェブサイトに掲載され、この記事はここに新しいタイトルの下に表示され、CC BY 4.0でライセンスを受けています。 TensorFlowについて TensorFlowについて