Support for DAG in unsupervised synthesis / imputation #93
+123
−6
Conversation
This file contains hidden or 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
|
|
- Fix density_() to use column_idx instead of hardcoded column 0 for target - Remove incorrect categorical_features index remapping during DAG processing When DAG reorders features for generation, the first feature to generate may not be the column at index 0. Previously, density_() always used column 0 as target regardless of which column was being generated, causing incorrect model training. Fixes synthetic data generation with DAG-based feature ordering.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.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.
Hi,
This PR is includes the support for causal DAGs in synthetic data generation/imputation.
DAGs are expressed as a dictionary
{ int: list[int] }where the key is the depenent column_idx and the value is a list of column indices the key depends on.E.g. for column 0 independent, column 1 depending on 0, column 2 depending on 1 and 0, you'd have:
The PR modifies (IMHO minimally)
src/tabpfn_extensions/unsupervised/unsupervised.pyand adds a respectiveexamples/unsupervised/generate_data_following_dag.py. Some small changes apply to other files.Regarding the
srcfile:impute_and order the variables using python'sgraphlib TopologicalSorterto synthesize them in the right order (e.g. independent first)conditional_idxof thecolumn_idxbeing generated/imputed to its dependencies (e.g., ifcolumn_idx==2thenconditional_idx==[0,1])What this PR does not do (based on the contribution guidelines):
unsupervisedto add to (am I missing something?). I hope the limited scope of the proposed changes + the provided example suffice. If you feel otherwise, it would be great if a test suite forunsupervisedis created, to which we could then add tests specific to the addition of dag support.