added new class SinkhornL1l2Transport() + dedicated test

Author: Slasnista

ot/da.py

@@ -1369,6 +1369,10 @@ class SinkhornLpl1Transport(BaseTransport):
 
     Parameters
     ----------
+    reg_e : float, optional (default=1)
+        Entropic regularization parameter
+    reg_cl : float, optional (default=0.1)
+        Class regularization parameter
     mode : string, optional (default="unsupervised")
         The DA mode. If "unsupervised" no target labels are taken into account
         to modify the cost matrix. If "semisupervised" the target labels
@@ -1384,6 +1388,11 @@ class SinkhornLpl1Transport(BaseTransport):
         The ground metric for the Wasserstein problem
     distribution : string, optional (default="uniform")
         The kind of distribution estimation to employ
+    max_iter : int, float, optional (default=10)
+        The minimum number of iteration before stopping the optimization
+        algorithm if no it has not converged
+    max_inner_iter : int, float, optional (default=200)
+        The number of iteration in the inner loop
     verbose : int, optional (default=0)
         Controls the verbosity of the optimization algorithm
     log : int, optional (default=0)
@@ -1452,3 +1461,103 @@ def fit(self, Xs, ys=None, Xt=None, yt=None):
             verbose=self.verbose, log=self.log)
 
         return self
+
+
+class SinkhornL1l2Transport(BaseTransport):
+    """Domain Adapatation OT method based on sinkhorn algorithm +
+    l1l2 class regularization.
+
+    Parameters
+    ----------
+    reg_e : float, optional (default=1)
+        Entropic regularization parameter
+    reg_cl : float, optional (default=0.1)
+        Class regularization parameter
+    mode : string, optional (default="unsupervised")
+        The DA mode. If "unsupervised" no target labels are taken into account
+        to modify the cost matrix. If "semisupervised" the target labels
+        are taken into account to set coefficients of the pairwise distance
+        matrix to 0 for row and columns indices that correspond to source and
+        target samples which share the same labels.
+    mapping : string, optional (default="barycentric")
+        The kind of mapping to apply to transport samples from a domain into
+        another one.
+        if "barycentric" only the samples used to estimate the coupling can
+        be transported from a domain to another one.
+    metric : string, optional (default="sqeuclidean")
+        The ground metric for the Wasserstein problem
+    distribution : string, optional (default="uniform")
+        The kind of distribution estimation to employ
+    max_iter : int, float, optional (default=10)
+        The minimum number of iteration before stopping the optimization
+        algorithm if no it has not converged
+    max_inner_iter : int, float, optional (default=200)
+        The number of iteration in the inner loop
+    verbose : int, optional (default=0)
+        Controls the verbosity of the optimization algorithm
+    log : int, optional (default=0)
+        Controls the logs of the optimization algorithm
+    Attributes
+    ----------
+    Coupling_ : the optimal coupling
+
+    References
+    ----------
+
+    .. [1] N. Courty; R. Flamary; D. Tuia; A. Rakotomamonjy,
+       "Optimal Transport for Domain Adaptation," in IEEE
+       Transactions on Pattern Analysis and Machine Intelligence ,
+       vol.PP, no.99, pp.1-1
+    .. [2] Rakotomamonjy, A., Flamary, R., & Courty, N. (2015).
+       Generalized conditional gradient: analysis of convergence
+       and applications. arXiv preprint arXiv:1510.06567.
+
+    """
+
+    def __init__(self, reg_e=1., reg_cl=0.1, mode="unsupervised",
+                 max_iter=10, max_inner_iter=200,
+                 tol=10e-9, verbose=False, log=False,
+                 metric="sqeuclidean",
+                 distribution_estimation=distribution_estimation_uniform,
+                 out_of_sample_map='ferradans'):
+
+        self.reg_e = reg_e
+        self.reg_cl = reg_cl
+        self.mode = mode
+        self.max_iter = max_iter
+        self.max_inner_iter = max_inner_iter
+        self.tol = tol
+        self.verbose = verbose
+        self.log = log
+        self.metric = metric
+        self.distribution_estimation = distribution_estimation
+        self.out_of_sample_map = out_of_sample_map
+
+    def fit(self, Xs, ys=None, Xt=None, yt=None):
+        """Build a coupling matrix from source and target sets of samples
+        (Xs, ys) and (Xt, yt)
+        Parameters
+        ----------
+        Xs : array-like of shape = [n_source_samples, n_features]
+            The training input samples.
+        ys : array-like, shape = [n_source_samples]
+            The class labels
+        Xt : array-like of shape = [n_target_samples, n_features]
+            The training input samples.
+        yt : array-like, shape = [n_labeled_target_samples]
+            The class labels
+        Returns
+        -------
+        self : object
+            Returns self.
+        """
+
+        super(SinkhornL1l2Transport, self).fit(Xs, ys, Xt, yt)
+
+        self.Coupling_ = sinkhorn_l1l2_gl(
+            a=self.mu_s, labels_a=ys, b=self.mu_t, M=self.Cost,
+            reg=self.reg_e, eta=self.reg_cl, numItermax=self.max_iter,
+            numInnerItermax=self.max_inner_iter, stopInnerThr=self.tol,
+            verbose=self.verbose, log=self.log)
+
+        return self

test/test_da.py

@@ -63,6 +63,56 @@ def test_sinkhorn_lpl1_transport_class():
     assert_equal(transp_Xs.shape, Xs.shape)
 
 
+def test_sinkhorn_l1l2_transport_class():
+    """test_sinkhorn_transport
+    """
+
+    ns = 150
+    nt = 200
+
+    Xs, ys = get_data_classif('3gauss', ns)
+    Xt, yt = get_data_classif('3gauss2', nt)
+
+    clf = ot.da.SinkhornL1l2Transport()
+
+    # test its computed
+    clf.fit(Xs=Xs, ys=ys, Xt=Xt)
+
+    # test dimensions of coupling
+    assert_equal(clf.Cost.shape, ((Xs.shape[0], Xt.shape[0])))
+    assert_equal(clf.Coupling_.shape, ((Xs.shape[0], Xt.shape[0])))
+
+    # test margin constraints
+    mu_s = unif(ns)
+    mu_t = unif(nt)
+    assert_allclose(np.sum(clf.Coupling_, axis=0), mu_t, rtol=1e-3, atol=1e-3)
+    assert_allclose(np.sum(clf.Coupling_, axis=1), mu_s, rtol=1e-3, atol=1e-3)
+
+    # test transform
+    transp_Xs = clf.transform(Xs=Xs)
+    assert_equal(transp_Xs.shape, Xs.shape)
+
+    Xs_new, _ = get_data_classif('3gauss', ns + 1)
+    transp_Xs_new = clf.transform(Xs_new)
+
+    # check that the oos method is not working
+    assert_equal(transp_Xs_new, Xs_new)
+
+    # test inverse transform
+    transp_Xt = clf.inverse_transform(Xt=Xt)
+    assert_equal(transp_Xt.shape, Xt.shape)
+
+    Xt_new, _ = get_data_classif('3gauss2', nt + 1)
+    transp_Xt_new = clf.inverse_transform(Xt=Xt_new)
+
+    # check that the oos method is not working and returns the input data
+    assert_equal(transp_Xt_new, Xt_new)
+
+    # test fit_transform
+    transp_Xs = clf.fit_transform(Xs=Xs, ys=ys, Xt=Xt)
+    assert_equal(transp_Xs.shape, Xs.shape)
+
+
 def test_sinkhorn_transport_class():
     """test_sinkhorn_transport
     """