iVAE.agent.agent

class iVAE.agent.agent(adata: AnnData, layer: str = 'counts', percent: float = 0.01, irecon: float = 0.0, beta: float = 1.0, dip: float = 0.0, tc: float = 0.0, info: float = 0.0, hidden_dim: int = 128, latent_dim: int = 10, i_dim: int = 2, lr: float = 0.0001, device: device = device(type='cpu'))

An agent class for modeling single-cell transcriptomics data using a variational autoencoder approach.

Parameters:

• adata (AnnData) – Annotated data matrix.

• layer (str, optional) – The layer of the AnnData object to use, by default ‘counts’.

• percent (float, optional) – Percent parameter value, by default 0.01.

• irecon (float, optional) – Irecon parameter value, by default 0.0.

• beta (float, optional) – Beta parameter value, by default 1.0.

• dip (float, optional) – Dip parameter value, by default 0.0.

• tc (float, optional) – TC parameter value, by default 0.0.

• info (float, optional) – Info parameter value, by default 0.0.

• hidden_dim (int, optional) – Hidden dimension size, by default 128.

• latent_dim (int, optional) – Latent dimension size, by default 10.

• i_dim (int, optional) – I dimension size, by default 2.

• lr (float, optional) – Learning rate, by default 1e-4.

• device (torch.device, optional) – Device to run the model on, by default uses GPU if available, otherwise CPU.

fit(epochs=1000)

Fits the model to the data for a specified number of epochs.

get_iembed()

Returns the intermediate embedding from the neural network.

get_latent()

Returns the latent representation of the data.

__init__(adata: AnnData, layer: str = 'counts', percent: float = 0.01, irecon: float = 0.0, beta: float = 1.0, dip: float = 0.0, tc: float = 0.0, info: float = 0.0, hidden_dim: int = 128, latent_dim: int = 10, i_dim: int = 2, lr: float = 0.0001, device: device = device(type='cpu'))

Methods

__init__(adata[, layer, percent, irecon, ...])
fit([epochs])	Fits the model to the data for a specified number of epochs.
get_iembed()	Returns the intermediate embedding from the neural network.
get_latent()	Returns the latent representation of the data.
load_data()
step(data)
take_latent(state)
update(states)