add sml/kmeans (#277)

按照最新的版本增加了kmeans (pr #235)

sml/kmeans/BUILD.bazel

@@ -0,0 +1,37 @@
+load("@rules_python//python:defs.bzl", "py_binary", "py_library", "py_test")
+
+package(default_visibility = ["//visibility:public"])
+
+py_library(
+    name = "kmeans",
+    srcs = ["kmeans.py"],
+    deps = [
+        "//sml/utils:fxp_approx",
+    ],
+)
+
+py_binary(
+    name = "kmeans_emul",
+    srcs = ["kmeans_emul.py"],
+    deps = [
+        ":kmeans",
+        "//examples/python/utils:dataset_utils",  # FIXME: remove examples dependency
+        "//sml/utils:emulation"
+    ],
+)
+
+
+
+py_test(
+    name = "kmeans_test",
+    srcs = ["kmeans_test.py"],
+    data = [
+        "//examples/python/conf",  # FIXME: remove examples dependency
+    ],
+    deps = [
+        ":kmeans",
+        "//examples/python/utils:dataset_utils",  # FIXME: remove examples dependency
+        "//spu:init",
+        "//spu/utils:simulation",
+    ],
+)

sml/kmeans/kmeans.py

@@ -0,0 +1,86 @@
+import jax
+import jax.numpy as jnp
+
+class KMEANS:
+    """
+    Parameters
+    ----------
+    n_clusters : int
+        The number of clusters to form as well as the number of
+        centroids to generate.
+
+    n_samples : int
+        The number of samples. 
+
+    max_iter : int, default=300
+        Maximum number of iterations of the k-means algorithm for a
+        single run.   
+    
+    tol : float, default=1e-4
+        Acceptable error to consider the two to be equal. 
+    """
+    def __init__(self, n_clusters, n_samples, max_iter=300, tol=1e-4):
+        self.n_clusters = n_clusters
+        self.max_iter = max_iter
+        self.tol = tol
+        self.init_params = jax.random.randint(jax.random.PRNGKey(1),shape=[self.n_clusters],minval=0,maxval=n_samples)
+        self._centers = jnp.zeros(())
+
+    def fit(self, x):
+        """Fit KMEANS.
+
+        Firstly, randomly select the initial centers. Then calculate the distance between each sample and each center, 
+        and assign each sample to the nearest center. Use an `aligned_array` to indicate the samples in a cluster, 
+        where unrelated samples will be set to 0. Once all samples are assigned, the center of each cluster will 
+        be updated to the average. The average could be got by `sum(data * aligned_array) / sum(aligned_array)`.
+        Different clusters could use broadcast for better performance.
+
+        Parameters
+        ----------
+        x : {array-like}, shape (n_samples, n_features)
+            Input data.
+
+        Returns
+        -------
+        self : object
+            Returns an instance of self.
+        """
+
+        centers = jnp.array([x[i] for i in self.init_params])
+        for _ in range(self.max_iter):
+            C = x.reshape((1, x.shape[0], x.shape[1])) - centers.reshape((centers.shape[0], 1, centers.shape[1]))
+            C = jnp.argmin(jnp.sum(jnp.square(C), axis=2), axis=0)
+
+            S = jnp.tile(C,(self.n_clusters,1))
+            ks = jnp.arange(self.n_clusters)
+            aligned_array_raw = (S.T - ks).T
+            aligned_array = jnp.equal(aligned_array_raw, 0)
+
+            centers_raw = x.reshape((1, x.shape[0], x.shape[1])) * aligned_array.reshape((aligned_array.shape[0],aligned_array.shape[1],1))
+            equals_sum = jnp.sum(aligned_array,axis=1)
+            centers_sum = jnp.sum(centers_raw, axis=1)
+            centers = jnp.divide(centers_sum.T, equals_sum).T 
+
+        self._centers = centers
+        return self
+
+    def predict(self, x):
+        """Result estimates.
+
+        Calculate the distance between each sample and each center, 
+        and assign each sample to the nearest center.
+
+        Parameters
+        ----------
+        x : {array-like}, shape (n_samples, n_features)
+            Input data for prediction.
+
+        Returns
+        -------
+        ndarray of shape (n_samples)
+            Returns the result of the sample for each class in the model.
+        """
+        centers = self._centers
+        y = x.reshape((1, x.shape[0], x.shape[1])) - centers.reshape((centers.shape[0], 1, centers.shape[1]))
+        y = jnp.argmin(jnp.sum(jnp.square(y), axis=2), axis=0)
+        return y

sml/kmeans/kmeans_emul.py

@@ -0,0 +1,54 @@
+import jax.numpy as jnp
+
+# from sklearn.metrics import roc_auc_score, explained_variance_score
+import sml.utils.emulation as emulation
+
+from sml.kmeans.kmeans import KMEANS
+from sklearn.datasets import make_blobs
+
+# TODO: design the enumation framework, just like py.unittest
+# all emulation action should begin with `emul_` (for reflection)
+def emul_KMEANS(mode: emulation.Mode.MULTIPROCESS):
+    def proc(x1, x2):
+        x = jnp.concatenate((x1, x2), axis=1)
+        model = KMEANS(
+            n_clusters=2,
+            n_samples=x.shape[0],
+            max_iter=10
+        )
+
+        return model.fit(x).predict(x)
+
+    def load_data():
+        n_samples = 1000
+        n_features = 100
+        X, _ = make_blobs(n_samples=n_samples,n_features=n_features,centers=2)
+        split_index = n_features//2
+        return X[:, :split_index], X[:, split_index:]
+
+    try:
+        # bandwidth and latency only work for docker mode
+        emulator = emulation.Emulator(
+            "examples/python/conf/3pc.json", mode, bandwidth=300, latency=20
+        )
+        emulator.up()
+
+        # load mock data
+        x1, x2  = load_data()
+        X = jnp.concatenate((x1, x2), axis=1)
+
+        # mark these data to be protected in SPU
+        x1, x2 = emulator.seal(x1, x2)
+        result = emulator.run(proc)(x1, x2)
+        print("result\n",result)
+
+        # Compare with sklearn
+        from sklearn.cluster import KMeans
+        model = KMeans(n_clusters=2)
+        print("sklearn:\n",model.fit(X).predict(X))
+    finally:
+        emulator.down()
+
+
+if __name__ == "__main__":
+    emul_KMEANS(emulation.Mode.MULTIPROCESS)

sml/kmeans/kmeans_test.py

@@ -0,0 +1,50 @@
+import unittest
+import json
+import jax.numpy as jnp
+
+# from sklearn.metrics import roc_auc_score, explained_variance_score
+import spu.utils.simulation as spsim
+import spu.spu_pb2 as spu_pb2  # type: ignore
+
+# TODO: unify this.
+import examples.python.utils.dataset_utils as dsutil
+
+from sml.kmeans.kmeans import KMEANS
+from sklearn.datasets import make_blobs
+
+class UnitTests(unittest.TestCase):
+    def test_kmeans(self):
+        sim = spsim.Simulator.simple(
+            3, spu_pb2.ProtocolKind.ABY3, spu_pb2.FieldType.FM64
+        )
+
+        def proc(x1, x2):
+            x = jnp.concatenate((x1, x2), axis=1)
+            model = KMEANS(
+                n_clusters=2,
+                n_samples=x.shape[0],
+                max_iter=10
+            )
+
+            return model.fit(x).predict(x)
+
+        def load_data():
+            n_samples = 1000
+            n_features = 100
+            X, _ = make_blobs(n_samples=n_samples,n_features=n_features,centers=2)
+            split_index = n_features//2
+            return X[:, :split_index], X[:, split_index:]
+
+        x1, x2 = load_data()
+        X = jnp.concatenate((x1, x2), axis=1)
+        result = spsim.sim_jax(sim, proc)(x1, x2)
+        print("result\n",result)
+
+        # Compare with sklearn
+        from sklearn.cluster import KMeans
+        model = KMeans(n_clusters=2)
+        print("sklearn:\n",model.fit(X).predict(X))
+
+
+if __name__ == "__main__":
+    unittest.main()