Meehai commited on
Commit
868825d
·
1 Parent(s): da3bc21

updated notebook

Browse files
data/train_set/.task_statistics.npz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8f601e71980200cacfc43a1b3ed5d7d932b538b8cf94d569d74911a0d3604b54
3
- size 28542
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f06911591fb45c4584e1fa8fcc099637d76cf6fac663a8400f6d76a2ac6ff97
3
+ size 29921
dronescapes_reader/multitask_dataset.py CHANGED
@@ -100,7 +100,6 @@ class MultiTaskDataset(Dataset):
100
  self._data_shape: tuple[int, ...] | None = None
101
  self._tasks: list[NpzRepresentation] | None = None
102
  self._default_vals: dict[str, tr.Tensor] | None = None
103
- self._name_to_task: dict[str, NpzRepresentation] | None = None
104
  if statistics is not None:
105
  self._statistics = self._load_external_statistics(statistics)
106
  else:
@@ -116,9 +115,7 @@ class MultiTaskDataset(Dataset):
116
 
117
  @property
118
  def name_to_task(self) -> dict[str, NpzRepresentation]:
119
- if self._name_to_task is None:
120
- self._name_to_task = {task.name: task for task in self.tasks}
121
- return self._name_to_task
122
 
123
  @property
124
  def default_vals(self) -> dict[str, tr.Tensor]:
@@ -195,6 +192,24 @@ class MultiTaskDataset(Dataset):
195
  raise e
196
  return res, items_name, self.task_names
197
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198
  # Private methods
199
 
200
  def _get_all_npz_files(self) -> dict[str, list[Path]]:
@@ -341,8 +356,11 @@ class MultiTaskDataset(Dataset):
341
  for task_name in self.task_names:
342
  task = [t for t in self.tasks if t.name == task_name][0]
343
  file_path = self.files_per_repr[task_name][index]
344
- file_path = None if file_path is None or not (fpr := file_path.resolve()).exists() else fpr
345
- res[task_name] = task.load_from_disk(file_path) if file_path is not None else self.default_vals[task_name]
 
 
 
346
  if not task.is_classification:
347
  if self.normalization is not None and self.normalization[task_name] == "min_max":
348
  res[task_name] = task.normalize(res[task_name])
 
100
  self._data_shape: tuple[int, ...] | None = None
101
  self._tasks: list[NpzRepresentation] | None = None
102
  self._default_vals: dict[str, tr.Tensor] | None = None
 
103
  if statistics is not None:
104
  self._statistics = self._load_external_statistics(statistics)
105
  else:
 
115
 
116
  @property
117
  def name_to_task(self) -> dict[str, NpzRepresentation]:
118
+ return {task.name: task for task in self.tasks}
 
 
119
 
120
  @property
121
  def default_vals(self) -> dict[str, tr.Tensor]:
 
192
  raise e
193
  return res, items_name, self.task_names
194
 
195
+ def add_task(self, task: NpzRepresentation):
196
+ """Safely adds a task to this reader. Most likely can be optimized"""
197
+ logger.debug(f"Adding a new task: '{task.name}'")
198
+ assert task.name not in self.task_names, f"Task '{task.name}' already exists: {self.task_names}"
199
+ self.task_names = sorted([*self.task_names, task.name])
200
+ self.task_types[task.name] = task
201
+ self._tasks = None
202
+ self.files_per_repr, self.file_names = self._build_dataset(self.task_types, self.task_names)
203
+
204
+ def remove_task(self, task_name: str):
205
+ """Safely removes a task from this reader"""
206
+ logger.debug(f"Removing a task: '{task_name}'")
207
+ assert task_name in self.task_names, f"Task '{task_name}' doesn't exist: {self.task_names}"
208
+ self.task_names = sorted(name for name in self.task_names if name != task_name)
209
+ del self.task_types[task_name]
210
+ self._tasks = None
211
+ self.files_per_repr, self.file_names = self._build_dataset(self.task_types, self.task_names)
212
+
213
  # Private methods
214
 
215
  def _get_all_npz_files(self) -> dict[str, list[Path]]:
 
356
  for task_name in self.task_names:
357
  task = [t for t in self.tasks if t.name == task_name][0]
358
  file_path = self.files_per_repr[task_name][index]
359
+ if file_path is None:
360
+ # TODO: I had a .resolve() and .exists() here? WTF? To fix in _build_dataset() maybe.
361
+ res[task_name] = self.default_vals[task_name]
362
+ else:
363
+ res[task_name] = task.load_from_disk(file_path)
364
  if not task.is_classification:
365
  if self.normalization is not None and self.normalization[task_name] == "min_max":
366
  res[task_name] = task.normalize(res[task_name])
scripts/dronescapes_viewer.ipynb CHANGED
The diff for this file is too large to render. See raw diff