add metatags for better SEO and change some wordings

Author: floriankozikowski

doc/api.rst

@@ -1,5 +1,8 @@
 .. _api_documentation:
 
+.. meta::
+   :description: Browse the skglm API documentation covering estimators (Lasso, ElasticNet, Cox), penalties (L1, SCAD, MCP), datafits (Logistic, Poisson), and optimized solvers.
+
 =================
 API Documentation
 =================

doc/contribute.rst

@@ -1,5 +1,9 @@
 .. _contribute:
 
+.. meta::
+   :description: Contribute to skglm by reporting bugs, suggesting features, or submitting pull requests. Join us in making skglm even better!
+    :og:title: Contribute to skglm
+
 Contribute to ``skglm``
 =======================
 

doc/getting_started.rst

@@ -1,5 +1,8 @@
 .. _getting_started:
 
+.. meta::
+   :description: Learn how to fit Lasso and custom GLM estimators with skglm, a modular Python library compatible with scikit-learn. Includes examples and code snippets.
+
 ===============
 Getting started
 ===============
@@ -31,7 +34,7 @@ Fitting a Lasso estimator
 -------------------------
 
 Let's start first by generating a toy dataset and splitting it to train and test sets.
-For that, we will use ``scikit-learn`` 
+For that, we will use ``scikit-learn``
 `make_regression <https://scikit-learn.org/stable/modules/generated/sklearn.datasets.make_regression.html#sklearn.datasets.make_regression>`_
 
 .. code-block:: python
@@ -42,7 +45,7 @@ For that, we will use ``scikit-learn``
 
     # generate toy data
     X, y = make_regression(n_samples=100, n_features=1000)
-    
+
     # split data
     X_train, X_test, y_train, y_test = train_test_split(X, y)
 
@@ -52,7 +55,7 @@ Then let's fit ``skglm`` :ref:`Lasso <skglm.Lasso>` estimator and prints its sco
 
     # import estimator
     from skglm import Lasso
-    
+
     # init and fit
     estimator = Lasso()
     estimator.fit(X_train, y_train)
@@ -63,7 +66,7 @@ Then let's fit ``skglm`` :ref:`Lasso <skglm.Lasso>` estimator and prints its sco
 
 .. note::
 
-    - The first fit after importing ``skglm`` has an overhead as ``skglm`` uses `Numba <https://numba.pydata.org/>`_ 
+    - The first fit after importing ``skglm`` has an overhead as ``skglm`` uses `Numba <https://numba.pydata.org/>`_
       The subsequent fits will achieve top speed since Numba compilation is cached.
 
 ``skglm`` has several other ``scikit-learn`` compatible estimators.
@@ -138,7 +141,7 @@ and pass it to ``GeneralizedLinearEstimator``. Explore the list of supported :re
 
 .. important::
 
-    - It is possible to create your own datafit and penalties. Check the tutorials on :ref:`how to add a custom datafit <how_to_add_custom_datafit>` 
+    - It is possible to create your own datafit and penalties. Check the tutorials on :ref:`how to add a custom datafit <how_to_add_custom_datafit>`
       and :ref:`how to add a custom penalty <how_to_add_custom_penalty>`.
 
 

doc/index.rst

@@ -3,17 +3,26 @@
    You can adapt this file completely to your liking, but it should at least
    contain the root `toctree` directive.
 
+.. meta::
+   :og:title: skglm: Fast, Scalable & Flexible Regularized GLMs and Sparse Modeling for Python
+   :description: skglm is the fastest, most modular Python library for regularized GLMs—fully scikit-learn compatible for advanced statistical modeling.
+   :og:image: _static/images/logo.svg
+   :og:url: https://contrib.scikit-learn.org/skglm/
+   :keywords: Generalized Linear Models, GLM, scikit-learn, Lasso, ElasticNet, Cox, modular, efficient, regularized
+
 =========
 ``skglm``
 =========
-*— A fast and modular scikit-learn replacement for regularized GLMs —*
+*— Fast and Flexible Generalized Linear Models for Python —*
 
 --------
 
 
 ``skglm`` is a Python package that offers **fast estimators** for regularized 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.
+that are **100% compatible with** ``scikit-learn``. It is **highly flexible** and supports a wide range of GLMs,
+designed to tackle high-dimensional data and scalable machine learning problems.
+Whether you choose from our ready-made estimators or **customize your own** using a modular combination of datafits and penalties,
+skglm delivers performance and flexibility for both academic research and production environments.
 
 Get a hands-on glimpse on ``skglm`` through the :ref:`Getting started page <getting_started>`.
 
@@ -79,6 +88,22 @@ It is also available on conda-forge and can be installed using, for instance:
     $ conda install -c conda-forge skglm
 
 With ``skglm`` being installed, Get the first steps with the package via the :ref:`Getting started section <getting_started>`.
+
+Applications and Use Cases
+---------------------------
+
+``skglm`` drives impactful solutions across diverse sectors with its fast, modular approach to regularized GLMs and sparse modeling. Some examples include:
+
+.. list-table::
+    :widths: 20 80
+
+    * - **Healthcare:**
+      - Enhance clinical trial analytics and early biomarker discovery by efficiently analyzing high-dimensional biological data and features like cox regression modeling.
+    * - **Finance:**
+      - Conduct transparent and interpretable risk modeling with scalable, robust sparse regression across vast datasets.
+    * - **Energy:**
+      - Optimize real-time electricity forecasting and load analysis by processing large time-series datasets for predictive maintenance and anomaly detection.
+
 Other advanced topics and uses-cases are covered in :ref:`Tutorials <tutorials>`.
 
 .. it is mandatory to keep the toctree here although it doesn't show up in the page

doc/tutorials/add_datafit.rst

@@ -1,6 +1,8 @@
 :orphan:
 
 .. _how_to_add_custom_datafit:
+.. meta::
+   :description: Tutorial on creating and implementing a custom datafit in skglm. Step-by-step guide includes deriving gradients, Hessians, and an example with Poisson datafit.
 
 How to add a custom datafit
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~

doc/tutorials/add_penalty.rst

@@ -2,6 +2,9 @@
 
 .. _how_to_add_custom_penalty:
 
+.. meta::
+   :description: Step-by-step tutorial on adding custom penalties in skglm. Covers implementation details, proximal operators, and optimality conditions using the L1 penalty.
+
 How to add a custom penalty
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

doc/tutorials/alpha_max.rst

@@ -1,5 +1,8 @@
 .. _alpha_max:
 
+.. meta::
+   :description: Tutorial explaining the critical regularization strength (alpha_max) in skglm. Learn conditions for zero solutions in L1-regularized optimization problems.
+
 ==========================================================
 Critical regularization strength above which solution is 0
 ==========================================================

doc/tutorials/cox_datafit.rst

@@ -1,10 +1,13 @@
 .. _maths_cox_datafit:
 
+.. meta::
+   :description: Detailed mathematical guide on Cox datafit implementation in skglm, covering Breslow and Efron estimates. Useful for survival analysis.
+
 =============================
 Mathematic behind Cox datafit
 =============================
 
-This tutorial presents the mathematics behind Cox datafit using both Breslow and Efron estimate. 
+This tutorial presents the mathematics behind Cox datafit using both Breslow and Efron estimate.
 
 
 Problem setup
@@ -95,7 +98,7 @@ where the division and the square operations are performed element-wise.
 The Hessian, as it is, is costly to evaluate because of the right hand-side term.
 In particular, the latter involves a :math:`\mathcal{O}(n^3)` operations. We overcome this limitation by using a diagonal upper bound on the Hessian.
 
-We construct such an upper bound by noticing that 
+We construct such an upper bound by noticing that
 
 #. the function :math:`F` is convex and hence :math:`\nabla^2 F(u)` is positive semi-definite
 #. the second term is positive semi-definite
@@ -127,10 +130,10 @@ We can define :math:`y_{i_1}, \ldots, y_{i_m}` the unique times, assumed to be i
 
 .. math::
     :label: def-H
-    
+
     H_{y_{i_l}} = \{ i \ | \ s_i = 1 \ ;\ y_i = y_{i_l} \}
     ,
-    
+
 the set of uncensored observations with the same time :math:`y_{i_l}`.
 
 Again, we refer to the expression of the negative log-likelihood according to Efron estimate [`2`_,  Section 6, equation (6.7)] to get the datafit formula
@@ -139,7 +142,7 @@ Again, we refer to the expression of the negative log-likelihood according to Ef
     :label: efron-estimate
 
     l(\beta) = \frac{1}{n} \sum_{l=1}^{m} (
-        \sum_{i \in H_{i_l}} - \langle x_i, \beta \rangle 
+        \sum_{i \in H_{i_l}} - \langle x_i, \beta \rangle
         + \sum_{i \in H_{i_l}} \log(\sum_{y_j \geq y_{i_l}} e^{\langle x_j, \beta \rangle} - \frac{\#(i) - 1}{ |H_{i_l} |}\sum_{j \in H_{i_l}} e^{\langle x_j, \beta \rangle}))
     ,
 
@@ -158,7 +161,7 @@ On the other hand, the minus term within :math:`\log` can be rewritten as a line
 
 .. math::
 
-    - \frac{\#(i) - 1}{| H_{i_l} |}\sum_{j \in H_{i_l}} e^{\langle x_j, \beta \rangle} 
+    - \frac{\#(i) - 1}{| H_{i_l} |}\sum_{j \in H_{i_l}} e^{\langle x_j, \beta \rangle}
         = \sum_{j=1}^{n} -\frac{\#(i) - 1}{| H_{i_l} |} \ \mathbb{1}_{j \in H_{i_l}} \ e^{\langle x_j, \beta \rangle}
         = \sum_{j=1}^n a_{i,j} e^{\langle x_j, \beta \rangle}
         = \langle a_i, e^{\mathbf{X}\beta} \rangle

doc/tutorials/intercept.rst

@@ -1,5 +1,8 @@
 .. _maths_unpenalized_intercept:
 
+.. meta::
+   :description: In-depth guide on intercept handling in skglm solvers. Covers mathematical derivations, gradient updates, Lipschitz constants, and examples for quadratic, logistic, and Huber datafits.
+
 Computation of the intercept
 ============================
 

doc/tutorials/prox_nn_group_lasso.rst

@@ -1,4 +1,6 @@
 .. _prox_nn_group_lasso:
+.. meta::
+   :description: Detailed tutorial on deriving the proximity operator and subdifferential for the positive group Lasso penalty in skglm. Includes mathematical proofs and examples.
 
 ===================================
 Details on the Positive Group Lasso