Update docs

Author: antoinedemathelin

adapt/feature_based/_cdan.py

@@ -34,8 +34,9 @@ def _get_default_classifier(name=None, state=None):
 @make_insert_doc(["encoder"])
 class CDAN(BaseAdaptDeep):
     """
-    CDAN (Conditional Adversarial Domain Adaptation) is an
-    unsupervised domain adaptation method on the model of the 
+    CDAN: Conditional Adversarial Domain Adaptation
+    
+    CDAN is an unsupervised domain adaptation method on the model of the 
     :ref:`DANN <adapt.feature_based.DANN>`. In CDAN the discriminator
     is conditioned on the prediction of the task network for
     source and target data. This should , in theory, focus the

adapt/feature_based/_mcd.py

@@ -14,8 +14,10 @@
 @make_insert_doc(["encoder", "task"])
 class MCD(BaseAdaptDeep):
     """
-    MCD: Maximum Classifier Discrepancy is a feature-based domain adaptation
-    method originally introduced for unsupervised classification DA.
+    MCD: Maximum Classifier Discrepancy
+    
+    MCD is a feature-based domain adaptation method originally introduced
+    for unsupervised classification DA.
     
     The goal of MCD is to find a new representation of the input features which
     minimizes the discrepancy between the source and target domains 

adapt/feature_based/_mdd.py

@@ -14,8 +14,10 @@
 @make_insert_doc(["encoder", "task"])
 class MDD(BaseAdaptDeep):
     """
-    MDD: Margin Disparity Discrepancy is a feature-based domain adaptation
-    method originally introduced for unsupervised classification DA.
+    MDD: Margin Disparity Discrepancy
+    
+    MDD is a feature-based domain adaptation method originally introduced
+    for unsupervised classification DA.
     
     The goal of MDD is to find a new representation of the input features which
     minimizes the disparity discrepancy between the source and target domains 

adapt/feature_based/_tca.py

@@ -50,7 +50,7 @@ class TCA(BaseAdaptEstimator):
     
     References
     ----------
-    .. [1] `[1] <https://www.cse.ust.hk/~qyang/Docs/2009/TCA.pdf>` S. J. Pan, \
+    .. [1] `[1] <https://www.cse.ust.hk/~qyang/Docs/2009/TCA.pdf>`_ S. J. Pan, \
 I. W. Tsang, J. T. Kwok and Q. Yang. "Domain Adaptation via Transfer Component \
 Analysis". In IEEE transactions on neural networks 2010
     """

adapt/feature_based/_wdgrl.py

@@ -13,8 +13,9 @@
 @make_insert_doc(["encoder", "task", "discriminator"])
 class WDGRL(BaseAdaptDeep):
     """
-    WDGRL (Wasserstein Distance Guided Representation Learning) is an
-    unsupervised domain adaptation method on the model of the 
+    WDGRL: Wasserstein Distance Guided Representation Learning
+    
+    WDGRL is an unsupervised domain adaptation method on the model of the 
     :ref:`DANN <adapt.feature_based.DANN>`. In WDGRL the discriminator
     is used to approximate the Wasserstein distance between the
     source and target encoded distributions in the spirit of WGAN.

adapt/instance_based/_balancedweighting.py

@@ -10,7 +10,7 @@ class BalancedWeighting(BaseAdaptEstimator):
     """
     BW : Balanced Weighting
     
-    Fit the estimator on source and target labeled data
+    Fit the estimator :math:`h` on source and target labeled data
     according to the modified loss:
     
     .. math::
@@ -26,7 +26,7 @@ class BalancedWeighting(BaseAdaptEstimator):
     
     Parameters
     ----------
-    gamma : float
+    gamma : float (default=0.5)
         ratio between 0 and 1 correspond to the importance
         given to the target labeled data. When `ratio=1`, the
         estimator is only fitted on target data. `ratio=0.5`

adapt/instance_based/_iwc.py

@@ -28,7 +28,7 @@ class IWC(BaseAdaptEstimator):
     
     .. math::
     
-    w(x) = \frac{1}{P(x \in Source)} - 1
+        w(x) = \\frac{1}{P(x \in Source)} - 1
 
     Parameters
     ----------

adapt/instance_based/_iwn.py

@@ -47,11 +47,12 @@ def weighted_mmd(Xs, Xt, weights, gamma=1.):
     return mxx + myy -2*mxy
 
 
-@make_insert_doc(["estimator", "weighter"], supervised=True)
+@make_insert_doc(["estimator", "weighter"])
 class IWN(BaseAdaptDeep):
     """
-    IWN : Importance Weighting Network is an instance-based method for
-    unsupervised domain adaptation<.
+    IWN : Importance Weighting Network
+    
+    IWN is an instance-based method for unsupervised domain adaptation.
     
     The goal of IWN is to reweight the source instances in order to
     minimize the Maximum Mean Discreancy (MMD) between the reweighted
@@ -107,7 +108,7 @@ class IWN(BaseAdaptDeep):
     
     References
     ----------
-    .. [1] `[1] <https://arxiv.org/pdf/2209.04215.pdff>`_ A. de Mathelin, F. Deheeger, \
+    .. [1] `[1] <https://arxiv.org/pdf/2209.04215.pdf>`_ A. de Mathelin, F. Deheeger, \
 M. Mougeot and N. Vayatis "Fast and Accurate Importance Weighting for \
 Correcting Sample Bias" In ECML-PKDD, 2022.
     """

adapt/instance_based/_rulsif.py

@@ -15,6 +15,8 @@
 
 EPS = np.finfo(float).eps
 
+
+@make_insert_doc()
 class RULSIF(BaseAdaptEstimator):
     """
     RULSIF: Relative Unconstrained Least-Squares Importance Fitting
@@ -45,15 +47,18 @@ class RULSIF(BaseAdaptEstimator):
     
     .. math::
     
-        \max_{\theta } \frac{1}{2}  \\theta^T H \\theta - h^T \\theta  + 
-        \frac{\\lambda}{2} \\theta^T \\theta
+        \max_{\\theta } \\frac{1}{2}  \\theta^T H \\theta - h^T \\theta  + 
+        \\frac{\\lambda}{2} \\theta^T \\theta
         
     where :
     
     .. math::
     
-        H_{ll'} = \\frac{\\alpha}{n_s} \sum_{x_i \\in X_S}  K(x_i, x_l) K(x_i, x_l') + \\frac{1-\\alpha}{n_t} \\sum_{x_i \\in X_T}  K(x_i, x_l) K(x_i, x_l') \\
-        h_{l}= \\frac{1}{n_T} \sum_{x_i \\in X_T} K(x_i, x_l)
+        H_{kl} = \\frac{\\alpha}{n_s} \sum_{x_i \\in X_S}  K(x_i, x_k) K(x_i, x_l) + \\frac{1-\\alpha}{n_T} \\sum_{x_i \\in X_T}  K(x_i, x_k) K(x_i, x_l)
+        
+    .. math::
+    
+        h_{k} = \\frac{1}{n_T} \sum_{x_i \\in X_T} K(x_i, x_k)
     
     The above OP is solved by the closed form expression
     
@@ -68,7 +73,7 @@ class RULSIF(BaseAdaptEstimator):
     
     .. math::
     
-        J = -\\frac{\\alpha}{2|X_S|} \\sum_{x \\in X_S} w(x)^2 - \frac{1-\\alpha}{2|X_T|} \\sum_{x \in X_T} w(x)^2
+        J = -\\frac{\\alpha}{2|X_S|} \\sum_{x \\in X_S} w(x)^2 - \\frac{1-\\alpha}{2|X_T|} \\sum_{x \in X_T} w(x)^2
     
     Finally, an estimator is fitted using the reweighted labeled source instances.
     

adapt/instance_based/_ulsif.py

@@ -44,22 +44,25 @@ class ULSIF(BaseAdaptEstimator):
     
     .. math::
     
-        \max_{\theta } \frac{1}{2}  \\theta^T H \\theta - h^T \\theta  + 
-        \frac{\\lambda}{2} \\theta^T \\theta
+        \max_{\\theta } \\frac{1}{2}  \\theta^T H \\theta - h^T \\theta  + 
+        \\frac{\\lambda}{2} \\theta^T \\theta
         
     where :
     
     .. math::
     
-        H_{ll'} = \\frac{1}{n_s} \sum_{x_i \\in X_S}  K(x_i, x_l) K(x_i, x_l') \\
-        h_{l}= \\frac{1}{n_T} \sum_{x_i \\in X_T} K(x_i, x_l)
+        H_{kl} = \\frac{1}{n_s} \sum_{x_i \\in X_S}  K(x_i, x_k) K(x_i, x_l)
+    
+    .. math::
+    
+        h_{k} = \\frac{1}{n_T} \sum_{x_i \\in X_T} K(x_i, x_k)
         
     
     The above OP is solved by the closed form expression
     
     .. math::
     
-        \hat{\\theta} = (H+\\lambda I_{n_s})^{(-1)} h 
+        \\hat{\\theta} = (H+\\lambda I_{n_s})^{(-1)} h 
     
     Furthemore the method admits a leave one out cross validation score that has a closed form expression 
     and can be used to select the appropriate parameters of the kernel function :math:`K` (typically, the paramter