0.11.0 #2354
lostella
announced in
Announcements
0.11.0
#2354
Replies: 1 comment
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Overview
Incremental training
Estimators are now re-trainable on new data, using the
train_from
method. This accepts a previously trained model (predictor), and new data to train on. The feature is integrated withgluonts.shell
-based SageMaker containers, and can be used by specifying the additionalmodel
channel to point to the output of a previous training job.New models
Two models are added in this release:
DeepVARHierarchicalEstimator
, a hierarchical extension toDeepVAREstimator
; learn more about how to use this in this tutorial.DeepNPTSEstimator
, a global extension to NPTS, where sampling probabilities are learned from data; learn more on how to use this estimator here.Removing deprecated import paths and options
This release removes the old, deprecated import paths for MXNet-based models. For example, now the MXNet-based
DeepAREstimator
will need to be imported fromgluonts.mx
(orgluonts.mx.model.deepar
).We also removed deprecated options for learning rate reduction in the
gluonts.mx.Trainer
class: these can now be controlled via theLearningRateReduction
callback.Dataset splitting functionality (experimental)
We updated the functionality to split time series datasets (along the time axis) for training/validation/test purposes. Now this functionality can be easily accessed via the
split
function (from gluonts.dataset.split import split
); learn more about this here.This feature is experimental and is subject to future changes.
Changelog
Breaking changes
(input, output)
pairs in the test split (Breaking: Update data splitters to return(input, output)
pairs in the test split #2031 by @npnv)mx.model
. (Breaking: Move MXNet-based models tomx.model
. #2126 by @Hongqing-work)mx.Trainer
. (Remove deprecated args frommx.Trainer
; rework LRS. #2153 by @jaheba)gluonts.core.exception
module. #2202 by @jaheba)gluonts.dataset.split
code, test, docs (Updategluonts.dataset.split
code, test, docs #2223 by @lostella)New features / major improvements
mx.Trainer
stops training. (Interruptingmx.Trainer
stops training. #2131 by @Hongqing-work)serde.dataclass
. (Addserde.dataclass
. #2166 by @jaheba)dataset.schema.translate
. (Add dataset.schema.translate. #2304 by @jaheba)forecast_start
to entry-wise metrics in evaluator (Addforecast_start
to entry-wise metrics in evaluator #2312 by @lostella)Bug fixes / minor improvements
DatasetCollection
for Python 3.9. #2135 by @rsnirwan)PandasDataset
for Python 3.9 (FixPandasDataset
for Python 3.9 #2141 by @lostella)PandasDataset
faster (MakePandasDataset
faster #2148 by @lostella)Timestamp
instead ofPeriod
(Fix Prophet wrapper to work withTimestamp
instead ofPeriod
#2182 by @lostella)gluonts.mx.batchify.stack
(Fix recursive case forgluonts.mx.batchify.stack
#2184 by @lostella)item_id
values inConstantValuePredictor
(Fixitem_id
values inConstantValuePredictor
#2192 by @codingWhale13)Patience
compatible withpydantic.dataclass
. #2197 by @jaheba)SymbolBlock
serde issue (FixSymbolBlock
serde issue #2187 by @lostella)r_forecast
wrapper to shift start date when truncating time series (Fixr_forecast
wrapper to shift start date when truncating time series #2216 by @abdulfatir)to_quantile_forecast
(Fix bug into_quantile_forecast
#2225 by @eugeneteoh)gluonts.mx.trainer.Trainer
in case of empty data loader (Fixgluonts.mx.trainer.Trainer
in case of empty data loader #2228 by @lostella)TimeSeriesSlice
performance (OptimizeTimeSeriesSlice
performance #2259 by @lostella)QuantileForecast.plot()
to useDateTimeIndex
(FixQuantileForecast.plot()
to useDateTimeIndex
#2269 by @abdulfatir)gluonts.dataset.split
for multivariate case (Fixgluonts.dataset.split
for multivariate case #2314 by @lostella)TestData
class ingluonts.dataset.split
(ImproveTestData
class ingluonts.dataset.split
#2315 by @lostella)make_evaluation_predictions
(Simplifymake_evaluation_predictions
#2309 by @lostella)kernel_size=1
(Fix MQCNN forkernel_size=1
#2321 by @lostella)BinnedUniforms
(Fix numerical bug inBinnedUniforms
#2344 by @moudheus)Documentation
installation
section. (Docs: Reworkinstallation
section. #2130 by @jaheba)DataFrame
example (Docs: Simplify wideDataFrame
example #2150 by @lostella)Test / setup changes
holidays
andmatplotlib
from core dependencies. (Removeholidays
andmatplotlib
from core dependencies. #2055 by @jaheba)This discussion was created from the release 0.11.0.
Beta Was this translation helpful? Give feedback.
All reactions