Releases: autogluon/fev
Releases Β· autogluon/fev
v0.5.0rc3
v0.5.0rc2
v0.5.0rc1
v0.4.1
v0.4.0
π¨ Highlights:
- The v0.4.0 release fixes an annoying error related to
dataset_fingerprint
validation infev.leaderboard
. We therefore highly recommend upgrading to the newest version.
Changes:
- Add support for rolling evaluation in
TaskGenerator
(#9)- Add new arguments
num_rolling_windows
,initial_cutoff
androlling_step_size
tofev.task.TaskGenerator
. These arguments make it possible to concisely define multiple backtesting splits from a single task.
- Add new arguments
- Make
dataset_fingerprint
depend only on the dataset contents (#10)- This PR ensures that the
dataset_fingerprint
is not affected by, e.g., changes to the dataset README on Hugging Face.
- This PR ensures that the
- Bump version to v0.4.0 (#11)
Full Changelog: v0.3.0...v0.4.0
v0.3.1
Changes:
- Rerun evaluations using the latest version of datasets on Hugging Face (v1.3.0). This includes the fix to the frequency of the
monash_hospital
anddominick
datasets - Remove
monash_saugeenday
dataset to fix the discrepancy between the Chronos paper and the leaderboard - Include results for TimesFM-2.0
- Make the error message caused by dataset fingerprint mismatch more informative
Full Changelog: v0.3.0...v0.3.1
v0.3.0
Changes:
- Add method
fev.utils.convert_long_df_to_hf_dataset
for converting long formatpd.DataFrame
datasets to Hugging Facedatasets.Dataset
format - Extend documentation for datasets, tasks, benchmarks and models
Full Changelog: v0.2.1...v0.3.0
v0.2.1
Changes:
- Remove harmonic mean aggregation from
fev.analysis.leaderboard()
Full Changelog: v0.2.0...v0.2.1
v0.2.0
Changes:
- Add a new column
trained_on_this_dataset
to the evaluation summary that tracks whether the model is used in zero-shot mode - Support new input formats for
leaderboard()
andpivot_table()
methods
Full Changelog: v0.1.0...v0.2.0