scglue.models.scglue.PairedSCGLUEModel.compile

PairedSCGLUEModel.compile(lam_data=1.0, lam_kl=1.0, lam_graph=0.02, lam_align=0.05, lam_sup=0.02, lam_joint_cross=0.02, lam_real_cross=0.02, lam_cos=0.02, dsc_steps=1, normalize_u=False, modality_weight=None, lr=2e-3, **kwargs)[source]

Prepare model for training

Parameters:

  • lam_data (float) – Data weight

  • lam_kl (float) – KL weight

  • lam_graph (float) – Graph weight

  • lam_align (float) – Adversarial alignment weight

  • lam_sup (float) – Cell type supervision weight

  • lam_joint_cross (float) – Joint cross-prediction weight

  • lam_real_cross (float) – Real cross-prediction weight

  • lam_cos (float) – Cosine similarity weight

  • dsc_steps (int) – Number of discriminator steps per encoder-decoder step

  • normalize_u (bool) – Whether to L2 normalize cell embeddings before decoder

  • modality_weight (typing.Optional[typing.Mapping[str, float]]) – Relative modality weight (indexed by modality name)

  • lr (float) – Learning rate

Return type:

None