draw_archs

behavenet.models.ae_model_architecture_generator.draw_archs(batch_size, input_dim, n_ae_latents, n_archs=100, check_memory=True, mem_limit_gb=5.0)[source]

Generate multiple random autoencoder architectures with a fixed number of latents.

Parameters:

  • batch_size (int) – expected batch size, to ensure that model and intermediate values will fit on gpu

  • input_dim (array-like) – dimensions of image with shape (n_channels, y_pix, x_pix)

  • n_ae_latents (int) – number of autoencoder latents - fixed for all generated architectures

  • n_archs (int, optional) – number of architectures to generate

  • check_memory (bool, optional) – True to check that the memory footprint of each architecture is below a certain threshold

  • mem_limit_gb (float, optional) – memory threshold in GB

Returns:

list of dicts, each of which fully defines a random architecture

Return type:

list of dict