an option to download only added files for a given dataset version #1100
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
I believe that this will also download the modified files (which is good), but maybe the name |
|
@eugen-ajechiloae-clearml done, I renamed it |
| unified_list = set() | ||
| for ds_id in datasets: | ||
| dataset = self.get(dataset_id=ds_id) | ||
| unified_list |= set(dataset._dataset_file_entries.values()) |
There was a problem hiding this comment.
@kirillfish This line seems to break:
unified_list |= set(dataset._dataset_file_entries.values())
TypeError: unhashable type: 'FileEntry'
Did you mean to construct the set based on the ids? Like:
unified_list |= {entry.id for entry in dataset._dataset_file_entries.values()}
There was a problem hiding this comment.
@kirillfish btw, here's a simple test snippet (you'll need to adapt it a little) to check that it works properly.
from pathlib import Path
import os
from clearml import Dataset, StorageManager
def main():
manager = StorageManager()
print("STEP1 : Downloading mnist dataset")
mnist_dataset = Path(manager.get_local_copy(
remote_url="https://allegro-datasets.s3.amazonaws.com/datasets/MNIST.zip", name="MNIST"))
mnist_dataset_train = mnist_dataset / "TRAIN"
mnist_dataset_test = mnist_dataset / "TEST"
print("STEP2 : Creating the training dataset")
mnist_dataset = Dataset.create(
dataset_project="dataset_examples", dataset_name="MNIST Training Dataset")
mnist_dataset.add_files(path=mnist_dataset_train, dataset_path="TRAIN")
mnist_dataset.upload()
mnist_dataset.finalize()
print("STEP3 : Create a child dataset of mnist dataset using TEST Dataset")
child_dataset = Dataset.create(
dataset_project="dataset_examples", dataset_name="MNIST Complete Dataset", parent_datasets=[mnist_dataset.id])
child_dataset.add_files(path=mnist_dataset_test, dataset_path="TEST")
child_dataset.upload()
child_dataset.finalize()
print("We are done, have a great day :)")
main()
dataset_path = Dataset.get("<replace with child dataset id, i.e. MNIST Complete Dataset ID>").get_local_copy()
print(Path(dataset_path).glob("*"))
os.rmtree(dataset_path)
dataset_path = Dataset.get("<replace with child dataset id, i.e. MNIST Complete Dataset ID>").get_local_copy(ignore_parent_datasets=True)
print(Path(dataset_path).glob("*"))
There was a problem hiding this comment.
@kirillfish any news on this?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
So far, when calling
get_local_copymethod on aclearml.Datasetobject, you would download all the files from the dataset AND all its parents recursively, or create soft links for all files that have been downloaded previously. But there was no way to get only files added to this particular version of the dataset, ignoring all the parents. This little PR implements this exact feature.Testing Instructions
parentdataset and add some fileschilddataset, inherited from the first one, and add some more filesonly_addedargument (Falseby default):data_base_dirwill contain only files returned bylist_added_filesOther Information
Potential issue:
Soft links are still being created for files in the diff which have already been downloaded - this is ok
But it you first call
child.get_local_copy(only_added=True)and then once againchild.get_local_copy(), it will not create soft links for existing files and download the diff once again -- probably not ok... The same applies to "grandchildren" datasets. Still figuring out why. On the other hand, this could be ok if we assumeonly_added=Trueflag is supposed to be used only for debug purposes or to quickly inspect datasets.