ModBetaGeoModel
Bayesian implementation of Modified BG/NBD model added.- Added
conditional_probability_of_being_alive_up_to_time
toParetoNBDFitter
. - Removed CSV model persistence and added a validation check for loading models from JSON files.
- Removed
dill
dependency for persistingfitters
models. - Converted
_check_inputs
into a static method and added a validation check for 'frequency < T'. - Converted
_dataframe_parser
from a static method and removed unnecessary_check_inputs
call. - Removed direct imports for
BaseModel
andPredictMixin
.
GammaGammaModel
Bayesian implementation of Gamma-Gamma model added.- Input validation added to
BaseModel
class. - Fixed array broadcasting bug in
BetaGeoModel
. - Revised frequency and monetary value descriptions in User Guide.
- Revised SQL code for monetary value calculations in documentation.
- Added required dependencies to
setup.cfg
.
- Bayesian predictions now supported, enabling entire probability distributions as well as point estimates for predictive outputs.
- Streamlined user API to minimize input arguments. All predictive methods are now also called from a single function.
model._idata
attribute now persisted as anarviz.InferenceData
object, and can be saved externally in JSON or CSV format. However, only JSONs can be loaded as of this release.- Documentation updated to latest versions of
sphinx
andpydata-sphinx-theme
. - Removed High Level Overview from documentation.
- Added deprecation warning for legacy Lifetimes
fitters
module. - Removed extraneous
lifetimes
import causing build issues. - Removed
psutils
library dependency. - Added
numpy >=1.20.0
library dependency. utils.posterior_predictive_deviation
metric removed pending further evaluation.- CI/CD pre-commit scripts added.
- Forked
lifetimes
library v0.11.3 and rebranded asbtyd
. BetaGeoCovarsFitter
BG/NBD model with time-invariant covariates added tofitters
module.- Alpha version of new modeling backend created in
models
module to support Bayesian modeling viapymc
. BetaGeoModel
Bayesian implementation of BG/NBD model added.- Switched to Apache 2.0 license.
- New experimental
posterior_predictive_deviation
metric added.