def visualize_model_performance(
output_dir: str,
domain_configs: List[DomainConfig],
dataset_types: List[str] = None,
device: str = "cuda:0",
reduction: str = "umap",
):
os.makedirs(output_dir, exist_ok=True)
if dataset_types is None:
dataset_types = ["train", "val"]
for dataset_type in dataset_types:
visualize_shared_latent_space(
domain_configs=domain_configs,
save_path=output_dir +
"/shared_latent_space_{}_{}.png".format(reduction, dataset_type),
dataset_type=dataset_type,
random_state=42,
reduction=reduction,
device=device,
)
# Todo resolve hacky try-catch way to avoid error if the domains consist of datasets of different sizes
try:
visualize_correlation_structure_latent_space(
domain_configs=domain_configs,
save_path=output_dir +
"/latent_space_correlation_structure_{}.png".format(
dataset_type),
dataset_type=dataset_type,
device=device,
)
except ValueError:
pass
for domain_config in domain_configs:
domain_name = domain_config.name
for dataset_type in dataset_types:
save_latents_to_csv(
domain_config=domain_config,
save_path=output_dir +
"/{}_latent_representations_{}.csv".format(
domain_name, dataset_type),
dataset_type=dataset_type,
device=device,
)
visualize_single_domain_cycling_correlation_structure(
domain_config=domain_config,
dataset_type=dataset_type,
save_path=output_dir +
"/{}_cycling_correlation_structure_{}.png".format(
domain_name, dataset_type),
device=device,
)
def save_latents_to_csv(
self,
domain_id: int = 0,
dataset_type: str = "val",
save_path: str = None,
posfix: str = "",
):
domain_config = self.domain_configs[domain_id]
if save_path is None:
save_path = os.path.join(
self.output_dir,
"latents_" + domain_config.name + "_" + dataset_type + posfix +
".csv",
)
save_latents_to_csv(
domain_config=domain_config,
dataset_type=dataset_type,
save_path=save_path,
device=self.device,
)
def visualize_model_performance(
output_dir: str,
domain_configs: List[DomainConfig],
dataset_types: List[str] = None,
device: str = "cuda:0",
):
os.makedirs(output_dir, exist_ok=True)
if dataset_types is None:
dataset_types = ["train", "val"]
for dataset_type in dataset_types:
visualize_shared_latent_space(
domain_configs=domain_configs,
save_path=output_dir
+ "/shared_latent_space_umap_{}.png".format(dataset_type),
dataset_type=dataset_type,
random_state=42,
reduction="umap",
device=device,
)
visualize_correlation_structure_latent_space(
domain_configs=domain_configs,
save_path=output_dir
+ "/latent_space_correlation_structure_{}.png".format(dataset_type),
dataset_type=dataset_type,
device=device,
)
for domain_config in domain_configs:
domain_name = domain_config.name
for dataset_type in dataset_types:
save_latents_to_csv(
domain_config=domain_config,
save_path=output_dir
+ "/{}_latent_representations_{}.csv".format(domain_name, dataset_type),
dataset_type=dataset_type,
device=device,
)