Changelog

Changelog#

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]#

Fixed#

  • ehrdata_blobs() now defaults layer to None and stores the generated time series tensor in .X. Previously it always wrote to .layers["tem_data"] regardless of the layer argument, so code that omitted layer silently analyzed a static 2D snapshot instead of the intended 3D time series. Pass layer= explicitly to keep the tensor in a named layer instead. (#284) @sueoglu

[0.4.0]#

Added#

Fixed#

  • harmonize_missing_values() and the read-time float casting no longer choke on boolean arrays. Boolean arrays are now treated like numeric ones (nothing to harmonize/cast). (#264) @eroell

  • move_to_obs() now reads values from the requested layer instead of always reading .X; passing layer= previously validated the layer but silently moved .X’s values. (#279) @eroell

  • gen_default_config() no longer fails with X must be 2-dimensional when the time series lives in a 3D .X (e.g. from physionet2019()); the selected source is reduced to the chosen timestep before writing, and layer now defaults to None (use .X). (#271) @eroell

  • infer_feature_types() binary detection contained a latent bug: the integrality guard used np.all(<generator>), which is always truthy and so never actually ran. The check is now the equivalent, correct set(col.unique()) == {0, 1}. No user-visible behaviour changes, as the disabled guard was redundant with the {0, 1} set check. (#268) @Zethson

  • infer_feature_types() no longer emits a warning with a blank feature name (Feature  was detected as categorical features stored numerically.) when no feature is uncertain. The warning is now only shown when at least one feature was detected as categorical stored numerically, and lists the affected feature names. (#267) @Zethson

  • Slicing a 3D X along the time axis now also slices .X. Previously, subsetting the third axis updated .shape and .tem but returned the parent’s full-length .X, so edata[:, :, idx].X.shape disagreed with edata[:, :, idx].shape. .X now applies the time-axis index exactly like a 3D layer. (#259) @eroell

Documentation#

  • The EHRData attribute reference now describes .obsm, .varm, .obsp and .varp in anndata’s terms instead of showing a generic placeholder. (#263) @eroell

  • Updated logo with 3D .X. Tutorials updated to use .X canonically. Docstring examples no longer pass layer="tem_data": the time series is kept in the default 3D .X. (#273) @eroell

[0.3.0]#

Added#

  • read_h5ed() and write_h5ed() are the new primary HDF5 I/O functions: .h5ed is now ehrdata’s on-disk format, marking it distinct from anndata’s .h5ad. read_h5ad and write_h5ad remain as deprecated aliases. (#252) @eroell

  • write_h5ed() and write_zarr() now write the ehrdata on-disk encoding 0.2.0. Because AnnData only guarantees 2D but not 3D arrays in X/layers (see scverse/anndata#2430), 3D arrays are relocated into .obsm (under reserved _ed_ondisk_X / _ed_ondisk_layers_<name> keys) and dropped from X/layers on write, then restored on read. Backwards compatibility for reading is maintained. The recommended file extensions for hdf5 are now .h5ed. (#252) @eroell

Fixed#

  • infer_feature_types() can handle EHRData objects with X as None. (#246) @sueoglu

  • physionet2019() no longer raises a shape mismatch on the full dataset: persons whose dynamic measurements all fall outside the observation window are now padded with missing values instead of being dropped from the time series tensor. (#251) @eroell

  • harmonize_missing_values() no longer logs a warning for each numeric layer when it has nothing to harmonize; reading a fully numeric dataset is now quiet.(#252) @eroell

Maintenance#

  • ehrdata is now compatible with anndata 0.13 (unified X/layers storage, IndexManager view indices, index type aliases moved to anndata.typing, and the stricter 2D-only X/layers on-disk rule) while remaining compatible with anndata <0.13. CI gains an integration-free core-anndata-min lane (pinned to anndata<0.13) and a pre-release lane to catch upstream breakage early. (#252) @eroell

  • DaskArray import fixed to follow anndata 0.12.16, and dask added to the intersphinx mapping. (#248) @sueoglu

  • CI now caches downloaded datasets used by ehrdata.dt to reduce flaky upstream hosts (e.g. physionet.org) breaking the test and notebook workflows. (#250) @eroell

  • Dataset downloads now use pooch instead of a custom requests-based downloader, aligning with the scverse ecosystem and providing caching out of the box. (#251) @eroell

  • The tqdm dependency has been removed. (#254) @eroell

Documentation#

[0.2.1]#

Fixed#

  • Compatibility with anndata>=0.12.13 (#240) @eroell

[0.2.0]#

Fixed#

  • Assigning .X to a view of an X-less EHRData (e.g. one created with layers= only) no longer raises TypeError: 'NoneType' object does not support item assignment. The view is now materialised before the assignment, consistent with how AnnData handles other field modifications on views. (#233) @eroell

Modified#

  • infer_feature_types() considers integers from 0, …, n as numeric. It further provides a new argument binary_as, to steer if columns 0/1 should be considered numeric or categorical. (#231) @eroell

[0.1.2]#

Added#

  • from_pandas() with format='long' provides a new keyword argument fill_time_gaps that fills missing timegaps in the common case of integer time steps from 0 to n_timesteps (#229) @eroell

Modified#

  • mimic_2() column censor_flg switched to lifeline’s convention with 1=event, 0=censored, before this dataset loader function had them vice versa since the dataset provides them as such originally. (#227) @sueoglu

Fixed#

  • from_pandas() with format='long' misordered entries in .X/.layers with .obs if the input df was not sorted for the obs id keys, which is now fixed. (#228) @eroell

Documentation#

  • Documentation style polishing (#223) @zethson

[0.1.1]#

Added#

Fixed#

  • Sliceing of EHRData objects fixed when the backing object is an AnnData. (#218) @eroell

Maintenance#

[0.1.0]#

Added#

Fixed#

  • read_h5ad fixed issues when backed=True. (#199) @eroell

  • read_h5ad fixed bug when .X is None and harmonize_missing_features is True. (#206) @eroell

  • setup_obs() with observation_table="person_visit_occurrence" now supports multiple visits per patient, creating one row per visit with unique observation IDs, instead of failing with xarray conversion errors with non-unique indices. (#210) @eroell

  • OMOP time interval boundaries now use half-open intervals [start, end) to prevent duplicate measurements at interval boundaries. (#210) @eroell

Maintenance#

  • Support Python3.14 (#194) @Zethson

  • Address FutureWarnings across multiple places (#200) @eroell

  • Enhanced tutorial structure (#208) @eroell

Modified#

  • Dataset generator function ed.dt.ehrdata_blobs now takes n_cat_var and n_categories arguments to generate categorical (integer encoded) time series data (#207) @sueoglu

  • If enrich_var_with_feature_info=True in setup_variables() and setup_interval_variables(), data_table_concept_ids not included within the concept table are now mapped from their respective alternate concept_id included in the concept_relationship table to retrieve the available feature information. (#205) @KilianDahm

  • setup_variables() and setup_interval_variables() with use of "person" now checks birth_datetime for meaningful behaviour and error messages. (#210) @eroell

  • gen_default_config() provides convenience to generate a config directly from an EHRData object, and should be used instead of the previous ehrdata.integrations.vitessce.gen_config. (#211) @eroell

[0.0.10]#

EHRData drops the .R field, and now supports 3D data storage in any slot of .layers. See the Getting started with EHRData tutorial for an introduction to this behaviour. In the future, .X will be enabled soon for 3D data storage as well.

Maintenance#

Fixed#

Modified#

  • EHRData drops the .R field in favor of using .layers for any 3D data arrays (#184) @eroell

  • EHRData’s shape property will always return a 3 dimensional shape. If an EHRData object has flat arrays only, the third dimension will be 1. (#184) @eroell

  • The following functions now take a layer argument: read_csv(), from_pandas(), to_pandas(), setup_variables(), setup_interval_variables(), ehrdata_blobs(), physionet2012(). If it is let to its default, None, the .X field of EHRData is used. Since .X is 2D in this release, in cases with 3D data, the layer argument needs to be used. (#184) @eroell

  • write_zarr() now writes an EHRData specific store encoding, with anndata as a substore. This change allows to use AnnData with its change to consolidated Zarr metadata, and better isolates AnnData’s io. (#185) @eroell

  • read_zarr() is adapted to read the new store encoding, and can also deal with AnnData stores. (#185) @eroell

[0.0.9]#

Maintenance#

  • Use custom logger & remove pydata sparse (#176) @Zethson

  • Replace figshare with scverse S3 (#177) @Zethson

  • Update template to v0.6.0 (#166) @Zethson

Fixed#

  • Fix order of var created in ed.io.omop.setup_variables and ed.io.omop.setup_interval_variables (#179) @eroell

Modified#

  • Rename ed.pl.vitessce.gen_config to ed.integrations.vitessce.gen_config (#181) @eroell

  • Rename ed.tl.omop.EHRDataset to ed.integrations.torch.OMOPEHRDataset (#181) @eroell

[0.0.8]#

Fixed#

  • Update duckdb imports for future (#157) @eroell

Maintenance#

  • Private subset method for EHRData (#160) @eroell

  • Remove omop package dependency (#160) @eroell

[0.0.7]#

Fixed#

  • Fix tests and Getting Started Notebook (#155) @eroell

Maintenance#

  • Update duckdb imports for future (#155) @eroell

[0.0.6]#

Fixed#

  • Cleaned up and updated tutorial notebooks (#140) @agerardy

Added#

[0.0.5]#

Fixed#

  • Initialize EHRData with X and layers (#132) @eroell

Added#

Modified#

  • Rename .t attribute to .tem

[0.0.4]#

Fixed#

  • Zarr version to less than 3

[0.0.3]#

Fixed#

  • Added missing zarr dependency

[0.0.2]#

Added#

  • Expanded documentation

  • Improved OMOP Extraction

  • Support for COO sparse matrices for R

  • A ed.dt.ehrdata_blobs test data generator function

  • Replace -1 encoded missing values with nans in physionet2012 challenge data

Breaking changes#

  • Renamed r to R

[0.0.1] - 2024-11-04#

Added#

  • Initial release

[Unreleased]#

Added#

  • Basic tool, preprocessing and plotting functions

Fixed#

  • tutorial notebooks updated to align with breaking changes