Conditional VAE

Reference:

Juntao Li et al. “Generating Classical Chinese Poems via Conditional Variational Autoencoder and Adversarial Training” in ACL 2018.

class textbox.model.VAE.cvae.CVAE(config, dataset)

Bases: Seq2SeqGenerator

We use the title of a poem and the previous line as condition to generate the current line.

forward(corpus, epoch_idx=0)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

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.

generate(batch_data, eval_data)

Predict the texts conditioned on a noise or sequence.

Parameters

• batch_data (Corpus) – Corpus class of a single batch.

• eval_data – Common data of all the batches.

Returns

Generated text, shape: [batch_size, max_len]

Return type

torch.Tensor

training: bool

xavier_uniform_initialization(module)

using uniform in PyTorch to initialize the parameters in nn.Embedding and nn.Linear layers. For bias in nn.Linear layers, using constant 0 to initialize.