doubleml-for-py

• Release highlight: Average potential outcomes for multiple discrete treatments
  via DoubleMLAPO and DoubleMLAPOS classes (proposed by Apoorva Lal) #245 #250

• Update User Guide and Example Gallery #188 #195

• Add sensitivity analysis to DoubleMLFramework #249

• Maintenance package #264 #265 #266

• Maintenance documentation #182 #184 #186 #193 #194 #196 #197

• API Update: Change nuisance evaluation for classifiers. The corresponding properties are renamed nuisance_loss instead of rmses #254 #184

• Add new example on sensitivity analysis #190

• Add a new example on DiD with DoubleML in R #178

• Enable set_sample_splitting for cluster data #255

• Update the make_confounded_irm_data data generating process #263

• Maintainance package #264

• Maintenance documentation #177 #180 #181 #187 #189

• Increment package requirements and update workflows for Python 3.9 (add tests for Python 3.12) #247 #175

• Additional example for ranking treatment effects (by Apoorva Lal) #173 #174

• Maintenance documentation #172

• Release highlight: Sample-selections models as DoubleMLSMM class (by Michaela Kecskésová) #231 #235 #171

• API change: Remove options apply_crossfitting and dml_procedure from the DoubleML class #227 #166

• Restructure the package to improve readability and maintainability #225

• Add a DoubleMLFramework class to combine multiple DoubleML models (aggregation of estimates, bootstrap, and CI-procedures #226 #169

• Enable the use of external predictions for short models in benchmarks (by Lucien) #238 #239

• Add the gain_statistics to utils for sensitivity analysis #229

• Maintenance documentation #162 #163 #164 #165 #167 #168

• Maintenance package #225 #229 #246

• Release highlight: Add weights to DoubleMLIRM class to extend sensitivity to GATEs etc. #220 #229 #155 #161

• Extend GATE and CATE estimation to the DoubleMLPLR class #220 #155

• Enable the use of external predictions for DoubleML classes #221 #159

• Implementing utility classes and functions (gain statistics and dummy learners) #221 #222 #229 #161

• Extend example Gallery #153 #158 #161

• Maintenance documentation #157 #160

• Maintenance package #223 #224

• Release highlight: Benchmarking for Sensitivity Analysis (omitted variable bias) #211

• Policy tree estimation for the DoubleMLIRM class #212

• Extending sensitivity and policy tree documentation in User Guide and Example Gallery #148 #150

• The package requirements are set to Python 3.8 or higher #211

• Maintenance documentation #149

• Maintenance package #213

• Fix install requirements for 0.6.2 #208

• Release highlight: Sensitivity Analysis (omitted variable bias) for #201

  • DoubleMLPLR
  • DoubleMLIRM
  • DoubleMLDID
  • DoubleMLDIDCS

• Updated documentation #144 #141

• Extend the guide with sensitivity and add further examples #142

• Maintenance package #202 #206

• Maintenance documentation #137 #138 #140 #143 #145 #146

DoubleML 0.6.1

• Release highlight: Difference-in-differences models for ATTE estimation #200 #194
  - Panel data DoubleMLDID
  - Repeated cross sections DoubleMLDIDCS

• Add a potential time variable to DoubleMLData (until now only used in DoubleMLDIDCS) #200

• Extend the guide in the documentation and add further examples #132 #133 #135

• Maintenance #199 #134 #136

DoubleML 0.6.0

• Release highlight: Heterogeneous treatment effects (GATE, CATE, Quantile effects, ...)

• Add out-of-sample RMSE and targets for nuisance elements and implement nuisance estimation
  evaluation via evaluate_learners(). #182 #188

• Implement gate() and cate() methods for DoubleMLIRM class. Both are
  based on the new DoubleMLBLP class. #169

• Implement different type of quantile models #179

  • Potential quantiles (PQ) in class DoubleMLPQ
  • Local potential quantiles (LPQ) in class DoubleMLLPQ
  • Conditional value at risk (CVaR) in class DoubleMLCVAR
  • Quantile treatment effects (QTE) in class DoubleMLQTE

• Extend clustering to nonlinear scores #190

• Add ipw_normalization option to DoubleMLIRM and DoubleMLIIVM #186

• Implement an abstract base class for data backends #173

• Code refactorings, bug fixes, docu updates, unit test extensions and continuous integration #183 #192 #195 #196

• Change License to BSD 3-Clause #198

• Maintenance #174 #178 #181

• Fix / adapted unit tests which failed in the release of 0.5.1 to conda-forge #172

• Store estimated models for nuisance parameters #159
• Bug fix: Overwrite for tune method (introduced for depreciation warning) did not return the tune result #160 #162
• Maintenance #166 #167 #168 #170

• Implement a new score function score = 'IV-type' for the PLIV model (for details see #151)
  --> API change from DoubleMLPLIV(obj_dml_data, ml_g, ml_m, ml_r [, ...]) to DoubleMLPLIV(obj_dml_data, ml_g, ml_m, ml_r, ml_g [, ...])
• Adapt the nuisance estimation for the 'IV-type' score for the PLR model (for details see #151)
  --> API change from DoubleMLPLR(obj_dml_data, ml_g, ml_m [, ...]) to DoubleMLPLR(obj_dml_data, ml_l, ml_m, ml_g [, ...])
• Allow the usage of classifiers for binary outcome variables in the model classes IRM and IIVM #134
• Published in JMLR: DoubleML - An Object-Oriented Implementation of Double Machine Learning in Python (citation info updated in #138 )
• Maintenance #143 #148 #149 #152 #153

• We added Contribution Guidelines, issue templates, a pull request template and a discussion forum to the repository #132
• Code refactorings, docu updates, unit test extensions and continuous integration #126 #127 #128 #130 #131

• Release highlight: Clustered standard errors for double machine learning models #116
• Improve exception handling for missings and infinite values in the confounders, predictions, etc. (fixes #120 by allowing null confounder values) #122
• Clean up dev requirements and use dev requirements on github actions #121
• Other updates #123

• Always use the same bootstrap algorithm independent of dml1 vs dml2 and consistent with docu and paper #101 & #102
• Added an exception handling to assure that an IV variable is specified when using a PLIV or IIVM model #107
• Improve exception handling for externally provided sample splitting #110
• Minor update of the str representation of DoubleMLData objects #112
• Code refactorings and unit test extensions #103, #105, #106, #111 & #113

• IIVM model: Added a subgroups option to adapt to cases with and without the subgroups of always-takers and never-takers (#96).
• Add checks for the intersections of y_col, d_cols, x_cols, z_cols (#84, #97). This also fixes #83 (with intersection between x_cols and d_cols a column could have been added multiple times to the covariate matrix).
• Added checks and exception handling for duplicate entries in d_cols, x_cols or z_cols (#100).
• Check the datatype of data when initializing DoubleMLData objects. Also check for duplicate column names (#100).
• Fix bug #95 in #97: It occurred when x_cols where inferred via setdiff and y_col was a string with multiple characters.
• We updated the citation info to refer to the arXiv paper (#98): Bach, P., Chernozhukov, V., Kurz, M. S., and Spindler, M. (2021), DoubleML - An Object-Oriented Implementation of Double Machine Learning in Python, arXiv:2104.03220.

• Provide an option to store & export the first-stage predictions #91
• Added the package logo to the doc

• Major extensions of the unit test framework which result in a coverage >98% (a summary is given in #82)
• In the PLR one can now also specify classifiers for ml_m in case of a binary treatment variable with values 0 and 1 (see #86 for details)
• The joint Python and R docu and user guide is now served to https://docs.doubleml.org from a separate repo https://github.com/DoubleML/doubleml-docs
• Generate and upload a unit test coverage report to codecov https://app.codecov.io/gh/DoubleML/doubleml-for-py #76
• Run lint checks with flake8 #78, align code with PEP8 standards #79, activate code quality checks at codacy #80
• Refactoring (reduce code redundancy) of the code for tuning of the ML learners used for approximation the nuisance functions #81
• Minor updates, bug fixes and improvements of the exception handling (contained in #82 & #89)

• Fixed a compatibility issue with scikit-learn 0.24, which only affected some unit tests (#70, #71)
• Added scheduled unit tests on github-action (three times a week) #69
• Split up estimation of nuisance functions and computation of score function components. Further introduced a private method _est_causal_pars_and_se(), see #72. This is needed for the DoubleML-Serverless project: https://github.com/DoubleML/doubleml-serverless.