23.8. d2l API ドキュメント

この節では、d2l パッケージに含まれるクラスと関数をアルファベット順に示し、それらが本書のどこで定義されているかを示す。これにより、より詳細な実装や説明を見つけることができる。 GitHub リポジトリ のソースコードも参照のこと。

23.8.1. クラス (Classes)

class d2l.torch.AdditiveAttention(num_hiddens, dropout, **kwargs)

ベースクラス: Module

Additive attention.

Defined in 11.3.2.2 章

forward(queries, keys, values, valid_lens)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.AddNorm(norm_shape, dropout)

ベースクラス: Module

The residual connection followed by layer normalization.

Defined in 11.7.2 章

forward(X, Y)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.AttentionDecoder

ベースクラス: Decoder

The base attention-based decoder interface.

Defined in 11.4 章

property attention_weights

class d2l.torch.Classifier(plot_train_per_epoch=2, plot_valid_per_epoch=1)

ベースクラス: Module

The base class of classification models.

Defined in 4.3 章

accuracy(Y_hat, Y, averaged=True)

Compute the number of correct predictions.

Defined in 4.3 章

layer_summary(X_shape)

Defined in 7.6 章

loss(Y_hat, Y, averaged=True)

Defined in 4.5 章

validation_step(batch)

class d2l.torch.DataModule(root='../data', num_workers=4)

ベースクラス: HyperParameters

The base class of data.

Defined in 3.2.2 章

get_dataloader(train)

get_tensorloader(tensors, train, indices=slice(0, None, None))

Defined in 3.3 章

train_dataloader()

val_dataloader()

class d2l.torch.Decoder

ベースクラス: Module

The base decoder interface for the encoder--decoder architecture.

Defined in 10.6 章

forward(X, state)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

init_state(enc_all_outputs, *args)

class d2l.torch.DotProductAttention(dropout)

ベースクラス: Module

Scaled dot product attention.

Defined in 11.3.2.2 章

forward(queries, keys, values, valid_lens=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.Encoder

ベースクラス: Module

The base encoder interface for the encoder--decoder architecture.

Defined in 10.6 章

forward(X, *args)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.EncoderDecoder(encoder, decoder)

ベースクラス: Classifier

The base class for the encoder--decoder architecture.

Defined in 10.6 章

forward(enc_X, dec_X, *args)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

predict_step(batch, device, num_steps, save_attention_weights=False)

Defined in 10.7.6 章

class d2l.torch.FashionMNIST(batch_size=64, resize=(28, 28))

ベースクラス: DataModule

The Fashion-MNIST dataset.

Defined in 4.2 章

get_dataloader(train)

Defined in 4.2 章

text_labels(indices)

Return text labels.

Defined in 4.2 章

visualize(batch, nrows=1, ncols=8, labels=[])

Defined in 4.2 章

class d2l.torch.GRU(num_inputs, num_hiddens, num_layers, dropout=0)

ベースクラス: RNN

The multilayer GRU model.

Defined in 10.3 章

class d2l.torch.HyperParameters

ベースクラス: object

The base class of hyperparameters.

save_hyperparameters(ignore=[])

Save function arguments into class attributes.

Defined in 23.7 章

class d2l.torch.LeNet(lr=0.1, num_classes=10)

ベースクラス: Classifier

The LeNet-5 model.

Defined in 7.6 章

class d2l.torch.LinearRegression(lr)

ベースクラス: Module

The linear regression model implemented with high-level APIs.

Defined in 3.5 章

configure_optimizers()

Defined in 3.5 章

forward(X)

Defined in 3.5 章

get_w_b()

Defined in 3.5 章

loss(y_hat, y)

Defined in 3.5 章

class d2l.torch.LinearRegressionScratch(num_inputs, lr, sigma=0.01)

ベースクラス: Module

The linear regression model implemented from scratch.

Defined in 3.4 章

configure_optimizers()

Defined in 3.4 章

forward(X)

Defined in 3.4 章

loss(y_hat, y)

Defined in 3.4 章

class d2l.torch.Module(plot_train_per_epoch=2, plot_valid_per_epoch=1)

ベースクラス: Module, HyperParameters

The base class of models.

Defined in 3.2 章

apply_init(inputs, init=None)

Defined in 6.4 章

configure_optimizers()

Defined in 4.3 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

loss(y_hat, y)

plot(key, value, train)

Plot a point in animation.

training_step(batch)

validation_step(batch)

class d2l.torch.MTFraEng(batch_size, num_steps=9, num_train=512, num_val=128)

ベースクラス: DataModule

The English-French dataset.

Defined in 10.5 章

build(src_sentences, tgt_sentences)

Defined in 10.5.3 章

get_dataloader(train)

Defined in 10.5.3 章

class d2l.torch.MultiHeadAttention(num_hiddens, num_heads, dropout, bias=False, **kwargs)

ベースクラス: Module

Multi-head attention.

Defined in 11.5 章

forward(queries, keys, values, valid_lens)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

transpose_output(X)

Reverse the operation of transpose_qkv.

Defined in 11.5 章

transpose_qkv(X)

Transposition for parallel computation of multiple attention heads.

Defined in 11.5 章

class d2l.torch.PositionalEncoding(num_hiddens, dropout, max_len=1000)

ベースクラス: Module

Positional encoding.

Defined in 11.6 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.PositionWiseFFN(ffn_num_hiddens, ffn_num_outputs)

ベースクラス: Module

The positionwise feed-forward network.

Defined in 11.7 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.ProgressBoard(xlabel=None, ylabel=None, xlim=None, ylim=None, xscale='linear', yscale='linear', ls=['-', '--', '-.', ':'], colors=['C0', 'C1', 'C2', 'C3'], fig=None, axes=None, figsize=(3.5, 2.5), display=True)

ベースクラス: HyperParameters

The board that plots data points in animation.

Defined in 3.2 章

draw(x, y, label, every_n=1)

Defined in 23.7 章

class d2l.torch.Residual(num_channels, use_1x1conv=False, strides=1)

ベースクラス: Module

The Residual block of ResNet models.

Defined in 8.6 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.ResNeXtBlock(num_channels, groups, bot_mul, use_1x1conv=False, strides=1)

ベースクラス: Module

The ResNeXt block.

Defined in 8.6.2 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.RNN(num_inputs, num_hiddens)

ベースクラス: Module

The RNN model implemented with high-level APIs.

Defined in 9.6 章

forward(inputs, H=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.RNNLM(rnn, vocab_size, lr=0.01)

ベースクラス: RNNLMScratch

The RNN-based language model implemented with high-level APIs.

Defined in 9.6 章

init_params()

output_layer(hiddens)

Defined in 9.5 章

class d2l.torch.RNNLMScratch(rnn, vocab_size, lr=0.01)

ベースクラス: Classifier

The RNN-based language model implemented from scratch.

Defined in 9.5 章

forward(X, state=None)

Defined in 9.5 章

init_params()

one_hot(X)

Defined in 9.5 章

output_layer(rnn_outputs)

Defined in 9.5 章

predict(prefix, num_preds, vocab, device=None)

Defined in 9.5 章

training_step(batch)

validation_step(batch)

class d2l.torch.RNNScratch(num_inputs, num_hiddens, sigma=0.01)

ベースクラス: Module

The RNN model implemented from scratch.

Defined in 9.5 章

forward(inputs, state=None)

Defined in 9.5 章

class d2l.torch.Seq2Seq(encoder, decoder, tgt_pad, lr)

ベースクラス: EncoderDecoder

The RNN encoder--decoder for sequence to sequence learning.

Defined in 10.7.3 章

configure_optimizers()

Defined in 4.3 章

validation_step(batch)

class d2l.torch.Seq2SeqEncoder(vocab_size, embed_size, num_hiddens, num_layers, dropout=0)

ベースクラス: Encoder

The RNN encoder for sequence-to-sequence learning.

Defined in 10.7 章

forward(X, *args)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.SGD(params, lr)

ベースクラス: HyperParameters

Minibatch stochastic gradient descent.

Defined in 3.4 章

step()

zero_grad()

class d2l.torch.SoftmaxRegression(num_outputs, lr)

ベースクラス: Classifier

The softmax regression model.

Defined in 4.5 章

forward(X)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.SyntheticRegressionData(w, b, noise=0.01, num_train=1000, num_val=1000, batch_size=32)

ベースクラス: DataModule

Synthetic data for linear regression.

Defined in 3.3 章

get_dataloader(train)

Defined in 3.3 章

class d2l.torch.TimeMachine(batch_size, num_steps, num_train=10000, num_val=5000)

ベースクラス: DataModule

The Time Machine dataset.

Defined in 9.2 章

build(raw_text, vocab=None)

Defined in 9.2 章

get_dataloader(train)

Defined in 9.3.3 章

class d2l.torch.Trainer(max_epochs, num_gpus=0, gradient_clip_val=0)

ベースクラス: HyperParameters

The base class for training models with data.

Defined in 3.2.2 章

clip_gradients(grad_clip_val, model)

Defined in 9.5 章

fit(model, data)

fit_epoch()

Defined in 3.4 章

prepare_batch(batch)

Defined in 6.7 章

prepare_data(data)

prepare_model(model)

Defined in 6.7 章

class d2l.torch.TransformerEncoder(vocab_size, num_hiddens, ffn_num_hiddens, num_heads, num_blks, dropout, use_bias=False)

ベースクラス: Encoder

The Transformer encoder.

Defined in 11.7.4 章

forward(X, valid_lens)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.TransformerEncoderBlock(num_hiddens, ffn_num_hiddens, num_heads, dropout, use_bias=False)

ベースクラス: Module

The Transformer encoder block.

Defined in 11.7.2 章

forward(X, valid_lens)

Define the computation performed at every call.

Should be overridden by all subclasses.

注釈

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class d2l.torch.Vocab(tokens=[], min_freq=0, reserved_tokens=[])

ベースクラス: object

Vocabulary for text.

to_tokens(indices)

property unk

23.8.2. 関数 (Functions)

d2l.torch.add_to_class(Class)

Register functions as methods in created class.

Defined in 3.2 章

d2l.torch.bleu(pred_seq, label_seq, k)

Compute the BLEU.

Defined in 10.7.6 章

d2l.torch.check_len(a, n)

Check the length of a list.

Defined in 9.5 章

d2l.torch.check_shape(a, shape)

Check the shape of a tensor.

Defined in 9.5 章

d2l.torch.corr2d(X, K)

Compute 2D cross-correlation.

Defined in 7.2 章

d2l.torch.cpu()

Get the CPU device.

Defined in 6.7 章

d2l.torch.gpu(i=0)

Get a GPU device.

Defined in 6.7 章

d2l.torch.init_cnn(module)

Initialize weights for CNNs.

Defined in 7.6 章

d2l.torch.init_seq2seq(module)

Initialize weights for sequence-to-sequence learning.

Defined in 10.7 章

d2l.torch.masked_softmax(X, valid_lens)

Perform softmax operation by masking elements on the last axis.

Defined in 11.3 章

d2l.torch.num_gpus()

Get the number of available GPUs.

Defined in 6.7 章

d2l.torch.plot(X, Y=None, xlabel=None, ylabel=None, legend=[], xlim=None, ylim=None, xscale='linear', yscale='linear', fmts=('-', 'm--', 'g-.', 'r:'), figsize=(3.5, 2.5), axes=None)

Plot data points.

Defined in 2.4 章

d2l.torch.set_axes(axes, xlabel, ylabel, xlim, ylim, xscale, yscale, legend)

Set the axes for matplotlib.

Defined in 2.4 章

d2l.torch.set_figsize(figsize=(3.5, 2.5))

Set the figure size for matplotlib.

Defined in 2.4 章

d2l.torch.show_heatmaps(matrices, xlabel, ylabel, titles=None, figsize=(2.5, 2.5), cmap='Reds')

Show heatmaps of matrices.

Defined in 11.1 章

d2l.torch.show_list_len_pair_hist(legend, xlabel, ylabel, xlist, ylist)

Plot the histogram for list length pairs.

Defined in 10.5 章

d2l.torch.try_all_gpus()

Return all available GPUs, or [cpu(),] if no GPU exists.

Defined in 6.7 章

d2l.torch.try_gpu(i=0)

Return gpu(i) if exists, otherwise return cpu().

Defined in 6.7 章

d2l.torch.use_svg_display()

Use the svg format to display a plot in Jupyter.

Defined in 2.4 章