MNT doc and whatsnew update pre 0.3 release (#181)

Author: Badr-MOUFAD

README.md

@@ -2,7 +2,7 @@
 
 # ``skglm``
 
-## A fast :zap: and modular :hammer_and_pick: scikit-learn replacement for sparse GLMs
+## A fast ⚡ and modular ⚒️ scikit-learn replacement for sparse GLMs
 
 </section>
 
@@ -13,9 +13,11 @@
 [![PyPI version](https://badge.fury.io/py/skglm.svg)](https://pypi.org/project/skglm/)
 
 
-``skglm`` is a Python package that offers **fast estimators** for sparse Generalized Linear Models (GLMs) that are **100% compatible with ``scikit-learn``**. It is **highly flexible** and supports a wide range of GLMs. You get to choose from ``skglm``'s already-made estimators or **customize your own** by combining the available datafits and penalty.
+``skglm`` is a Python package that offers **fast estimators** for sparse Generalized Linear Models (GLMs) that are **100% compatible with ``scikit-learn``**. It is **highly flexible** and supports a wide range of GLMs.
+You get to choose from ``skglm``'s already-made estimators or **customize your own** by combining the available datafits and penalties.
+
+Excited to have a tour on ``skglm`` [documentation](https://contrib.scikit-learn.org/skglm/)?
 
-Excited to have a tour on ``skglm`` [documentation](https://contrib.scikit-learn.org/skglm/) :memo:?
 
 # Why ``skglm``?
 
@@ -25,10 +27,10 @@ There are several reasons to opt for ``skglm`` among which:
 
 |  |  |
 | ----- | -------------- |
-| **Speed** :zap: | Fast solvers able to tackle large datasets, either dense or sparse, with millions of features **up to 100 times faster** than ``scikit-learn``|
-| **Modularity** :hammer_and_pick: | User-friendly API than enables **composing custom estimators** with any combination of its existing datafits and penalties |
-| **Extensibility** :arrow_up_down: | Flexible design that makes it **simple and easy to implement new datafits and penalties**, a matter of few lines of code
-| **Compatibility** :electric_plug: | Estimators **fully compatible with the ``scikit-learn`` API** and drop-in replacements of its GLM estimators
+| **Speed** | Fast solvers able to tackle large datasets, either dense or sparse, with millions of features **up to 100 times faster** than ``scikit-learn``|
+| **Modularity** | User-friendly API that enables **composing custom estimators** with any combination of its existing datafits and penalties |
+| **Extensibility** | Flexible design that makes it **simple and easy to implement new datafits and penalties**, a matter of few lines of code
+| **Compatibility** | Estimators **fully compatible with the ``scikit-learn`` API** and drop-in replacements of its GLM estimators
 |  |  |
 
 
@@ -70,7 +72,7 @@ print(estimator.score(X, y))
 ```
 You can refer to the documentation to explore the list of ``skglm``'s already-made estimators.
 
-Didn't find one that suits you :monocle_face:, you can still compose your own.
+Didn't find one that suits you? you can still compose your own.
 Here is a code snippet that fits a MCP-regularized problem with Huber loss.
 
 ```python

doc/api.rst

@@ -41,6 +41,7 @@ Penalties
    L2
    L2_3
    MCPenalty
+   PositiveConstraint
    WeightedL1
    WeightedGroupL2
    SCAD
@@ -93,6 +94,8 @@ Experimental
 .. autosummary::
    :toctree: generated/
 
-   SqrtLasso
-   PDCD_WS
-   IterativeReweightedL1
+   IterativeReweightedL1
+   PDCD_WS 
+   Pinball
+   SqrtQuadratic
+   SqrtLasso

doc/changes/0.3.rst

@@ -0,0 +1,16 @@
+.. _changes_0_3:
+
+Version 0.3 (2023/07/01)
+------------------------
+
+- Add :ref:`Cox Estimator <skglm.CoxEstimator>` with support of L1, L2, and Elastic regularization (PR: :gh:`171`)
+- Reduce time overhead when fitting :ref:`Lasso Estimator <skglm.Lasso>` (PR: :gh:`129`)
+- Add :ref:`Cox <skglm.datafits.Cox>` datafit for survival analysis (PR :gh:`180`, :gh:`168`, :gh:`159`, :gh:`157`)
+- Experimental :ref:`Pinball <skglm.experimental.Pinball>` datafit (PR: :gh:`134`)
+- Add :ref:`Gamma <skglm.datafits.Gamma>` datafit (PR: :gh:`113`)
+- Add Positivity constraint to :ref:`L1 <skglm.penalties.L1>`, :ref:`L1_plus_L2 <skglm.penalties.L1_plus_L2>`, :ref:`WeightedL1 <skglm.penalties.WeightedL1>` (PR: :gh:`110`)
+- Add :ref:`PositiveConstraint <skglm.penalties.PositiveConstraint>` (PR: :gh:`126`)
+- Add :ref:`L-BFGS <skglm.solvers.LBFGS>` solver for problems with smooth datafits and penalties (PR: :gh:`165`, :gh:`173`)
+- Experimental :ref:`Primal-dual coordinate descent solve <skglm.experimental.PDCD_WS>` for problems with non-smooth datafits (PR: :gh:`131`)
+- Add support of ``float32`` in :ref:`ProxNewton <skglm.solvers.ProxNewton>`  
+ 

skglm/experimental/__init__.py

@@ -1,9 +1,12 @@
 from .reweighted import IterativeReweightedL1
-from .sqrt_lasso import SqrtLasso
+from .sqrt_lasso import SqrtLasso, SqrtQuadratic
 from .pdcd_ws import PDCD_WS
+from .quantile_regression import Pinball
 
 __all__ = [
-    SqrtLasso,
     IterativeReweightedL1,
-    PDCD_WS
+    PDCD_WS,
+    Pinball,
+    SqrtQuadratic,
+    SqrtLasso,
 ]

skglm/experimental/sqrt_lasso.py

@@ -12,9 +12,11 @@
 
 
 class SqrtQuadratic(BaseDatafit):
-    """Unnormalized square root quadratic datafit.
+    r"""Unnormalized square root quadratic datafit.
 
-    The datafit reads::
+    The datafit reads:
+
+    .. math::
 
         ||y - Xw||_2
     """