Dev

bio_vae/datasets.py

@@ -12,13 +12,16 @@
 
 from functools import lru_cache
 
+from albumentations import Compose
+from typing import Callable
+
 
 class DatasetGlob(Dataset):
  def __init__(
  self,
  path_glob,
  over_sampling=1,
- transform=None,
+ transform: Callable = Compose([]),
  samples=-1,
  shuffle=True,
  **kwargs,

bio_vae/lightning/dataloader.py

@@ -42,25 +42,35 @@ def __init__(
  def get_dataset(self):
  return self.dataset
 
- def splitting(self, dataset, split=0.8, seed=42):
- if len(dataset) < 4:
- return dataset, dataset, dataset, dataset
- spliting_shares = [
- len(dataset) * split * split, # train
- len(dataset) * split * (1 - split), # test
- len(dataset) * split * (1 - split), # predict
- len(dataset) * (1 - split) * (1 - split), # val
- ]
-
- train, test, predict, val = random_split(
+
+ def splitting(self, dataset, split_train=0.8, split_val=0.1, seed=42):
+ if len(dataset) < 3:
+ return dataset, dataset, dataset
+
+ train_share = int(len(dataset) * split_train)
+ val_share = int(len(dataset) * split_val)
+ test_share = len(dataset) - train_share - val_share
+
+ # Ensure that the splits add up correctly
+ if train_share + val_share + test_share != len(dataset):
+ raise ValueError("The splitting ratios do not add up to the length of the dataset")
+
+ torch.manual_seed(seed) # for reproducibility
+
+ train, val, test = random_split(
  dataset,
- list(map(int, saferound(spliting_shares, places=0))),
+ [train_share, val_share, test_share]
  )
 
- return test, train, predict, val
+ return train, val, test
 
  def setup(self, stage=None):
- self.test, self.train, self.predict, self.val = self.splitting(self.dataset)
+ self.train, self.val, self.test = self.splitting(self.dataset)
+
+ # self.test = self.get_dataloader(test)
+ # self.predict = self.get_dataloader(predict)
+ # self.train = self.get_dataloader(train)
+ # self.val = self.get_dataloader(val)
 
  def test_dataloader(self):
  return DataLoader(self.test, **self.data_loader_settings)
@@ -72,7 +82,7 @@ def val_dataloader(self):
  return DataLoader(self.val, **self.data_loader_settings)
 
  def predict_dataloader(self):
- return DataLoader(self.predict, **self.data_loader_settings)
+ return DataLoader(self.dataset, **self.data_loader_settings)
 
  # def teardown(self, stage: Optional[str] = None):
  # # Used to clean-up when the run is finished

environment.yml

@@ -2,8 +2,6 @@
 channels:
  - bioconda
  - pytorch
- - idr
- - ome
  - conda-forge
  # - defaults
  - torch

pyproject.toml

@@ -28,7 +28,7 @@ umap-learn = {extras = ["plot"], version = "^0.5.3"}
 colorcet = "^3.0.1"
 holoviews = "^1.15.2"
 # idr-py = "^0.4.2"
-llvmlite = "^0.39.1"
+#llvmlite = "^0.39.1"
 torchmetrics = "^0.11.0"
 tensorboard = "^2.11.2"
 albumentations = "^1.3.0"