scmg.model.manifold_generation module

scmg.model.manifold_generation.sinusoidal_embedding(n, d)

Create sinusoidal embeddings.

class scmg.model.manifold_generation.ConditionalDiffusionModel(n_feature, n_time_feature, condition_classes, n_condition_feature, n_steps=1000, min_beta=0.0001, max_beta=0.02, n_network_blocks=3)

Bases: Module

A denoise diffusion model.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x0, t, eta=None)

Add noise to the data. x0 has shape (n_batch, n_feature) and t has shape (n_batch,).

backward(x, t, x_cond)

Predict the mean of distribution at t-1 given the data at t.

generate(x_cond)

Generate samples from the model.

class scmg.model.manifold_generation.RecurrentBlock(n_input, n_output, n_hidden)

Bases: Module

A MLP style recurrent block.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x)

Define 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.

class scmg.model.manifold_generation.MLPDenoiser(n_feature, n_time_feature, n_condition_feature, n_hidden=2048, n_blocks=3)

Bases: Module

A simple MLP denoiser.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x, t_emb, x_cond)

Define 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.

scmg.model.manifold_generation.train_diffusion_model(model, loader, num_epochs, output_path, lr=0.0001, loss_history=None)

Train the diffusion model.

scmg.model.manifold_generation.generate_cells(model, cond_classes, batch_size=512)

Generate cells from the model.

scmg.model.manifold_generation.generate_transition_cells(model, start_cell_type, end_cell_type, n_cells, batch_size=512)

Generate cells between two cell types.