bug fixes

deeprvat/deeprvat/train.py

@@ -29,7 +29,7 @@
     PearsonCorrTorch,
     RSquared,
 )
-from deeprvat.utils import suggest_hparams
+from deeprvat.utils import resolve_path_with_env, suggest_hparams
 from numcodecs import Blosc
 from pytorch_lightning.callbacks import ModelCheckpoint
 from pytorch_lightning.callbacks.early_stopping import EarlyStopping
@@ -226,6 +226,7 @@ def __init__(
         split: str = "train",
         cache_tensors: bool = False,
         temp_dir: Optional[str] = None,
+        chunksize: int = 1000,
         # samples: Optional[Union[slice, np.ndarray]] = None,
         # genes: Optional[Union[slice, np.ndarray]] = None
     ):
@@ -239,6 +240,7 @@ def __init__(
         )
 
         self.cache_tensors = cache_tensors
+        self.chunksize = chunksize
         if self.cache_tensors:
             logger.info("Keeping all input tensors in main memory")
 
@@ -258,10 +260,11 @@ def __init__(
             pheno: pheno_data["samples"][split]
             for pheno, pheno_data in self.data.items()
         }
-        if temp_dir is not None:
-            self.input_tensor_dir = temp_dir
-        else:
-            self.input_tensor_dir = TemporaryDirectory(prefix="training_data", dir=".")
+        temp_path = (Path(resolve_path_with_env(temp_dir)) / "deeprvat_training"
+                  if temp_dir is not None
+                  else Path("deeprvat_training"))
+        temp_path.mkdir(parents=True, exist_ok=True)
+        self.input_tensor_dir = TemporaryDirectory(prefix="training_data", dir=str(temp_path))
 
         self.subset_samples()
 
@@ -353,7 +356,7 @@ def subset_samples(self):
                 zarr.save_array(
                     tensor_path,
                     pheno_data["input_tensor_zarr"][:][self.samples[pheno]],
-                    chunks=(1000, None, None, None),
+                    chunks=(self.chunksize, None, None, None),
                     compressor=Blosc(clevel=1),
                 )
                 pheno_data["input_tensor_zarr"] = zarr.open(tensor_path)
@@ -363,7 +366,7 @@ def subset_samples(self):
                 f"{n_samples_orig} samples kept"
             )
 
-    def index_input_tensor_zarr(pheno: str, indices: np.ndarray):
+    def index_input_tensor_zarr(self, pheno: str, indices: np.ndarray):
         # IMPORTANT!!! Never call this function after self.subset_samples()
 
         x = self.data[pheno]["input_tensor_zarr"]
@@ -388,6 +391,8 @@ def __init__(
         num_workers: Optional[int] = 0,
         pin_memory: bool = False,
         cache_tensors: bool = False,
+        temp_dir: Optional[str] = None,
+        chunksize: int = 1000,
     ):
         logger.info("Intializing datamodule")
 
@@ -451,6 +456,8 @@ def __init__(
             "num_workers",
             "pin_memory",
             "cache_tensors",
+            "temp_dir",
+            "chunksize",
         )
 
     def upsample(self) -> np.ndarray:
@@ -488,6 +495,8 @@ def train_dataloader(self):
             self.hparams.batch_size,
             split="train",
             cache_tensors=self.hparams.cache_tensors,
+            temp_dir=self.hparams.temp_dir,
+            chunksize=self.hparams.chunksize,
         )
         return DataLoader(
             dataset,
@@ -507,6 +516,8 @@ def val_dataloader(self):
             self.hparams.batch_size,
             split="val",
             cache_tensors=self.hparams.cache_tensors,
+            temp_dir=self.hparams.temp_dir,
+            chunksize=self.hparams.chunksize,
         )
         return DataLoader(
             dataset,
@@ -565,6 +576,8 @@ def run_bagging(
                 "upsampling_factor",
                 "sample_with_replacement",
                 "cache_tensors",
+                "temp_dir",
+                "chunksize",
             )
         }
         dm = MultiphenoBaggingData(