cleaned typing

src/fdiff/dataloaders/datamodules.py

@@ -30,8 +30,8 @@ def __init__(
         self.X = X
         self.y = y
         self.standardize = standardize
-        self.feature_mean = None
-        self.feature_std = None
+        self.feature_mean = torch.empty(size=(self.X.size(1), self.X.size(2)))
+        self.feature_std = torch.empty(size=(self.X.size(1), self.X.size(2)))
         if X_ref is None:
             X_ref = X
         self.compute_feature_statistics(X_ref)

src/fdiff/models/score_models.py

@@ -99,7 +99,7 @@ def forward(self, batch: DiffusableBatch) -> torch.Tensor:
 
     def training_step(
         self, batch: DiffusableBatch, batch_idx: int, dataloader_idx: int = 0
-    ) -> torch.Tensor:
+    ):
         loss = self.training_loss_fn(self, batch)
 
         self.log_dict(
@@ -135,8 +135,9 @@ def configure_optimizers(self) -> OptimizerLRScheduler:
 
     def set_loss_fn(self) -> tuple[Callable, Callable]:
         # depending on the scheduler, get the right loss function
+        scheduler_config = self.noise_scheduler.config  # type: ignore
         if isinstance(self.noise_scheduler, DDPMScheduler):
-            self.max_time = self.noise_scheduler.config.num_train_timesteps
+            self.max_time = scheduler_config.num_train_timesteps
 
             training_loss_fn = get_ddpm_loss(
                 scheduler=self.noise_scheduler, train=True, max_time=self.max_time
@@ -167,7 +168,7 @@ def set_loss_fn(self) -> tuple[Callable, Callable]:
                 "Scheduler not implemented yet, cannot set loss function"
             )
 
-    def set_time_encoder(self) -> nn.Module:
+    def set_time_encoder(self):
         if isinstance(self.noise_scheduler, DDPMScheduler):
             return TimeEncoding(d_model=self.d_model, max_time=self.max_time)
 

src/fdiff/models/transformer.py

@@ -68,7 +68,6 @@ def __init__(self, d_model: int, scale=30.0):
             torch.randn((d_model + 1) // 2) * scale, requires_grad=False
         )
         self.dense = nn.Linear(d_model, d_model)
-        print(self.d_model)
 
     def forward(self, x: torch.Tensor, timesteps: torch.Tensor) -> torch.Tensor:
         time_proj = timesteps[:, None] * self.W[None, :] * 2 * np.pi
@@ -78,4 +77,6 @@ def forward(self, x: torch.Tensor, timesteps: torch.Tensor) -> torch.Tensor:
 
         t_emb = t_emb.unsqueeze(1)
 
-        return x + self.dense(t_emb)
+        projected_emb: torch.Tensor = self.dense(t_emb)
+
+        return x + projected_emb

src/fdiff/schedulers/custom_ddpm_scheduler.py

@@ -1,3 +1,5 @@
+# mypy: ignore-errors
+
 # Copyright 2023 UC Berkeley Team and The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

src/fdiff/schedulers/vpsde_scheduler.py

@@ -71,9 +71,7 @@ def add_noise(
         sample = mean + noise
         return sample
 
-    def get_beta(
-        self, timestep: torch.Tensor | float | np.ndarray
-    ) -> torch.Tensor | float | np.ndarray:
+    def get_beta(self, timestep: torch.Tensor | float | np.ndarray) -> torch.Tensor:
         return torch.tensor(
             self.beta_0 + timestep * (self.beta_1 - self.beta_0), device=self.device
         )

src/fdiff/utils/losses.py

@@ -33,7 +33,7 @@ def get_sde_loss_fn(
         else lambda *args, **kwargs: 0.5 * torch.sum(*args, **kwargs)
     )
 
-    def loss_fn(model, batch):
+    def loss_fn(model, batch) -> torch.Tensor:
         """Compute the loss function.
 
         Args:
@@ -118,7 +118,7 @@ def loss_fn(model, batch):
 
 
 def get_ddpm_loss(scheduler, train, max_time):
-    def loss_fn(model, batch):
+    def loss_fn(model, batch) -> torch.Tensor:
         if train:
             model.train()
         else: