[TO-REVIEW] Add multi-domain Monge alignment and JCPOT Target shift m…

…ethod (#180) * add per domaij split * better tets multi_domains * first shit multimonge alignment * add test * add exmaple * update maping to new API * add proper references * add JCPOT * add stuff * add jcpot tests * upate doc and add JCPOT * exmale in gallery of jcpot * typo in exmaple
scikit-adaptation · Jul 19, 2024 · ce1d60c · ce1d60c
1 parent fb26ff5
commit ce1d60c
Show file tree

Hide file tree

Showing 10 changed files with 849 additions and 6 deletions.
diff --git a/README.md b/README.md
@@ -228,3 +228,10 @@ The library is distributed under the 3-Clause BSD license.
 [27] S. Si, D. Tao and B. Geng. In IEEE Transactions on Knowledge and Data Engineering, (2010) [Bregman Divergence-Based Regularization for Transfer Subspace Learning](https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=4118b4fc7d61068b9b448fd499876d139baeec81)
 
 [28] Solomon, J., Rustamov, R., Guibas, L., & Butscher, A. (2014, January). [Wasserstein propagation for semi-supervised learning](https://proceedings.mlr.press/v32/solomon14.pdf). In International Conference on Machine Learning (pp. 306-314). PMLR.
+
+[29] Montesuma, Eduardo Fernandes, and Fred Maurice Ngole Mboula. ["Wasserstein barycenter for multi-source domain adaptation."](https://openaccess.thecvf.com/content/CVPR2021/papers/Montesuma_Wasserstein_Barycenter_for_Multi-Source_Domain_Adaptation_CVPR_2021_paper.pdf) In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 16785-16793. 2021.
+
+[30] Gnassounou, Theo, Rémi Flamary, and Alexandre Gramfort. ["Convolution Monge Mapping Normalization for learning on sleep data."](https://proceedings.neurips.cc/paper_files/paper/2023/file/21718991f6acf19a42376b5c7a8668c5-Paper-Conference.pdf) Advances in Neural Information Processing Systems 36 (2024).
+
+[31] Redko, Ievgen, Nicolas Courty, Rémi Flamary, and Devis Tuia.[ "Optimal transport for multi-source domain adaptation under target shift."](https://proceedings.mlr.press/v89/redko19a/redko19a.pdf) In The 22nd International Conference on artificial intelligence and statistics, pp. 849-858. PMLR, 2019.
+
diff --git a/examples/methods/plot_label_prop_da.py b/examples/methods/plot_label_prop_da.py
@@ -31,8 +31,14 @@
 from sklearn.kernel_ridge import KernelRidge
 from sklearn.linear_model import LogisticRegression
 from sklearn.metrics import mean_squared_error
+from sklearn.svm import SVC
 
-from skada import OTLabelPropAdapter, make_da_pipeline, source_target_split
+from skada import (
+    JCPOTLabelPropAdapter,
+    OTLabelPropAdapter,
+    make_da_pipeline,
+    source_target_split,
+)
 from skada.datasets import make_shifted_datasets
 
 # %%
@@ -330,3 +336,96 @@
 
 plt.title("Propagated labels data")
 plt.axis(ax)
+
+
+# %%
+# Generate classification classification dataset and plot it
+# -----------------------------------------------------
+#
+# We generate a simple 2D target shift dataset.
+
+X, y, sample_domain = make_shifted_datasets(
+    n_samples_source=20,
+    n_samples_target=20,
+    shift="target_shift",
+    noise=0.2,
+    random_state=42,
+)
+
+
+Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
+
+
+plt.figure(5, (10, 5))
+plt.subplot(1, 2, 1)
+plt.scatter(Xs[:, 0], Xs[:, 1], c=ys, cmap="tab10", vmax=9, label="Source")
+plt.title("Source data")
+ax = plt.axis()
+
+plt.subplot(1, 2, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Target")
+plt.title("Target data")
+plt.axis(ax)
+
+
+# %%
+# Train with LabelProp and JCPOT + classifier
+# -------------------------
+#
+# On this target shift dataset, we can see that the label propagation method
+# does not work well because it finds correspondences between the source and
+# target samples with different classes. In this case JCPOT is more robust
+# to this kind of shift because it estimates the class proportions in the target.
+
+
+clf = make_da_pipeline(OTLabelPropAdapter(), SVC())
+clf.fit(X, y, sample_domain=sample_domain)
+
+clf_jcpot = make_da_pipeline(JCPOTLabelPropAdapter(reg=0.1), SVC())
+clf_jcpot.fit(X, y, sample_domain=sample_domain)
+
+
+yt_pred = clf.predict(Xt)
+acc_t = (yt_pred == yt).mean()
+
+print(f"LabelProp Accuracy on target: {acc_t:.2f}")
+
+
+yt_pred = clf_jcpot.predict(Xt)
+acc_s_jcpot = (yt_pred == yt).mean()
+
+print(f"JCPOT Accuracy on target: {acc_s_jcpot:.2f}")
+
+XX, YY = np.meshgrid(np.linspace(ax[0], ax[1], 100), np.linspace(ax[2], ax[3], 100))
+Z = clf.predict(np.c_[XX.ravel(), YY.ravel()]).reshape(XX.shape)
+Z_jcpot = clf_jcpot.predict(np.c_[XX.ravel(), YY.ravel()]).reshape(XX.shape)
+
+
+plt.figure(7, (10, 5))
+
+
+plt.subplot(1, 2, 1)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Prediction")
+plt.imshow(
+    Z,
+    extent=(ax[0], ax[1], ax[2], ax[3]),
+    origin="lower",
+    alpha=0.5,
+    cmap="tab10",
+    vmax=9,
+)
+plt.title(f"LabelProp reglog on target (ACC={acc_t:.2f})")
+plt.axis(ax)
+
+plt.subplot(1, 2, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Prediction")
+plt.imshow(
+    Z_jcpot,
+    extent=(ax[0], ax[1], ax[2], ax[3]),
+    origin="lower",
+    alpha=0.5,
+    cmap="tab10",
+    vmax=9,
+)
+plt.title(f"JCPOT reglog on target (ACC={acc_s_jcpot:.2f})")
+plt.axis(ax)
diff --git a/examples/methods/plot_monge_alignment_da.py b/examples/methods/plot_monge_alignment_da.py
@@ -0,0 +1,259 @@
+"""
+Multi-domain Linear Monge Alignment
+===================================
+
+This example illustrates the use of the MultiLinearMongeAlignmentAdapter
+
+"""
+
+# Author: Remi Flamary
+#
+# License: BSD 3-Clause
+# sphinx_gallery_thumbnail_number = 4
+
+# %% Imports
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.linear_model import LogisticRegression
+
+from skada import (
+    MultiLinearMongeAlignmentAdapter,
+    make_da_pipeline,
+    source_target_split,
+)
+from skada.datasets import make_shifted_datasets
+
+# %%
+# Generate concept drift classification dataset and plot it
+# -----------------------------------------------------
+#
+# We generate a simple 2D concept drift dataset.
+
+X, y, sample_domain = make_shifted_datasets(
+    n_samples_source=20,
+    n_samples_target=20,
+    shift="concept_drift",
+    noise=0.2,
+    label="multiclass",
+    random_state=42,
+)
+
+
+Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
+
+
+plt.figure(5, (10, 5))
+plt.subplot(1, 2, 1)
+plt.scatter(Xs[:, 0], Xs[:, 1], c=ys, cmap="tab10", vmax=9, label="Source")
+plt.title("Source data")
+ax = plt.axis()
+
+plt.subplot(1, 2, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Target")
+plt.title("Target data")
+plt.axis(ax)
+
+# %%
+# Train a classifier on source data
+# --------------------------------
+#
+# We train a simple SVC classifier on the source domain and evaluate its
+# performance on the source and target domain. Performance is much lower on
+# the target domain due to the shift. We also plot the decision boundary
+
+
+clf = MultiLinearMongeAlignmentAdapter()
+clf.fit(X, sample_domain=sample_domain)
+
+X_adapt = clf.transform(X, sample_domain=sample_domain, allow_source=True)
+
+
+plt.figure(5, (10, 3))
+plt.subplot(1, 3, 1)
+plt.scatter(Xs[:, 0], Xs[:, 1], c=ys, cmap="tab10", vmax=9, label="Source")
+plt.title("Source data")
+ax = plt.axis()
+
+plt.subplot(1, 3, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Target")
+plt.title("Target data")
+plt.axis(ax)
+
+plt.subplot(1, 3, 3)
+plt.scatter(
+    X_adapt[sample_domain >= 0, 0],
+    X_adapt[sample_domain >= 0, 1],
+    c=y[sample_domain >= 0],
+    marker="o",
+    cmap="tab10",
+    vmax=9,
+    label="Source",
+    alpha=0.5,
+)
+plt.scatter(
+    X_adapt[sample_domain < 0, 0],
+    X_adapt[sample_domain < 0, 1],
+    c=y[sample_domain < 0],
+    marker="x",
+    cmap="tab10",
+    vmax=9,
+    label="Target",
+    alpha=1,
+)
+plt.legend()
+plt.title("Adapted data")
+
+
+# %%
+# Train a classifier on adapted data
+# ----------------------------------
+
+clf = make_da_pipeline(
+    MultiLinearMongeAlignmentAdapter(),
+    LogisticRegression(),
+)
+
+clf.fit(X, y, sample_domain=sample_domain)
+
+print(
+    "Average accuracy on all domains:",
+    clf.score(X, y, sample_domain=sample_domain, allow_source=True),
+)
+
+# %% Multisource and taregt data
+
+
+def get_multidomain_data(
+    n_samples_source=100,
+    n_samples_target=100,
+    noise=0.1,
+    random_state=None,
+    n_sources=3,
+    n_targets=2,
+):
+    np.random.seed(random_state)
+    X, y, sample_domain = make_shifted_datasets(
+        n_samples_source=n_samples_source,
+        n_samples_target=n_samples_target,
+        noise=noise,
+        shift="concept_drift",
+        label="multiclass",
+        random_state=random_state,
+    )
+    for ns in range(n_sources - 1):
+        Xi, yi, sample_domaini = make_shifted_datasets(
+            n_samples_source=n_samples_source,
+            n_samples_target=n_samples_target,
+            noise=noise,
+            shift="concept_drift",
+            label="multiclass",
+            random_state=random_state + ns,
+            mean=np.random.randn(2),
+            sigma=np.random.rand(2) * 0.5 + 0.5,
+        )
+        Xs, Xt, ys, yt = source_target_split(Xi, yi, sample_domain=sample_domaini)
+        X = np.vstack([X, Xt])
+        y = np.hstack([y, yt])
+        sample_domain = np.hstack([sample_domain, np.ones(Xt.shape[0]) * (ns + 2)])
+
+    for nt in range(n_targets - 1):
+        Xi, yi, sample_domaini = make_shifted_datasets(
+            n_samples_source=n_samples_source,
+            n_samples_target=n_samples_target,
+            noise=noise,
+            shift="concept_drift",
+            label="multiclass",
+            random_state=random_state + nt + 42,
+            mean=np.random.randn(2),
+            sigma=np.random.rand(2) * 0.5 + 0.5,
+        )
+        Xs, Xt, ys, yt = source_target_split(Xi, yi, sample_domain=sample_domaini)
+        X = np.vstack([X, Xt])
+        y = np.hstack([y, yt])
+        sample_domain = np.hstack([sample_domain, -np.ones(Xt.shape[0]) * (nt + 1)])
+
+    return X, y, sample_domain
+
+
+X, y, sample_domain = get_multidomain_data(
+    n_samples_source=50,
+    n_samples_target=50,
+    noise=0.1,
+    random_state=43,
+    n_sources=3,
+    n_targets=2,
+)
+
+Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
+
+
+plt.figure(5, (10, 5))
+plt.subplot(1, 2, 1)
+plt.scatter(Xs[:, 0], Xs[:, 1], c=ys, cmap="tab10", vmax=9, label="Source")
+plt.title("Source data")
+ax = plt.axis()
+
+plt.subplot(1, 2, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Target")
+plt.title("Target domains")
+plt.axis(ax)
+
+
+# %%
+clf = MultiLinearMongeAlignmentAdapter()
+clf.fit(X, sample_domain=sample_domain)
+
+X_adapt = clf.transform(X, sample_domain=sample_domain, allow_source=True)
+
+
+plt.figure(5, (10, 3))
+plt.subplot(1, 3, 1)
+plt.scatter(Xs[:, 0], Xs[:, 1], c=ys, cmap="tab10", vmax=9, label="Source")
+plt.title("Source data")
+ax = plt.axis()
+
+plt.subplot(1, 3, 2)
+plt.scatter(Xt[:, 0], Xt[:, 1], c=yt, cmap="tab10", vmax=9, label="Target")
+plt.title("Target data")
+plt.axis(ax)
+
+plt.subplot(1, 3, 3)
+plt.scatter(
+    X_adapt[sample_domain >= 0, 0],
+    X_adapt[sample_domain >= 0, 1],
+    c=y[sample_domain >= 0],
+    marker="o",
+    cmap="tab10",
+    vmax=9,
+    label="Source",
+    alpha=0.5,
+)
+plt.scatter(
+    X_adapt[sample_domain < 0, 0],
+    X_adapt[sample_domain < 0, 1],
+    c=y[sample_domain < 0],
+    marker="x",
+    cmap="tab10",
+    vmax=9,
+    label="Target",
+    alpha=1,
+)
+plt.legend()
+plt.axis(ax)
+plt.title("Adapted data")
+
+# %%
+# Train a classifier on adapted data
+# ----------------------------------
+
+clf = make_da_pipeline(
+    MultiLinearMongeAlignmentAdapter(),
+    LogisticRegression(),
+)
+
+clf.fit(X, y, sample_domain=sample_domain)
+
+print(
+    "Average accuracy on all domains:",
+    clf.score(X, y, sample_domain=sample_domain, allow_source=True),
+)