def _initialize(self):
"""Set the initial weights, means and covs (with full covariance matrix).
weights: the prior of the clusters (what percentage of data does a cluster have)
means: the mean points of the clusters
covs: the covariance matrix of the clusters
"""
self.weights = np.ones(self.K)
if self.init == 'random':
self.means = [
self.X[x] for x in random.sample(range(self.n_samples), self.K)
]
self.covs = [np.cov(self.X.T) for _ in range(K)]
elif self.init == 'kmeans':
kmeans = KMeans(K=self.K, max_iters=self.max_iters // 3, init='++')
kmeans.fit(self.X)
self.assignments = kmeans.predict()
self.means = kmeans.centroids
self.covs = []
for i in np.unique(self.assignments):
self.weights[int(i)] = (self.assignments == i).sum()
self.covs.append(np.cov(self.X[self.assignments == i].T))
else:
raise ValueError('Unknown type of init parameter')
self.weights /= self.weights.sum()
def kmeans_example(plot=False):
X, y = make_blobs(centers=4, n_samples=500, n_features=2,
shuffle=True, random_state=42)
clusters = len(np.unique(y))
k = KMeans(K=clusters, max_iters=150, init='++')
k.fit(X)
k.predict()
if plot:
k.plot()
def robust_example(plot=False):
X, y = load_robust()
clusters = 5
k = KMeans(K=clusters, max_iters=50, init="++")
k.fit(X)
k.predict()
if plot:
k.plot()
def _initialize(self):
"""Set the initial weights, means and covs (with full covariance matrix).
weights: the prior of the clusters (what percentage of data does a cluster have)
means: the mean points of the clusters
covs: the covariance matrix of the clusters
"""
self.weights = np.ones(self.K)
if self.init == 'random':
self.means = [self.X[x] for x in random.sample(range(self.n_samples), self.K)]
self.covs = [np.cov(self.X.T) for _ in range(K)]
elif self.init == 'kmeans':
kmeans = KMeans(K=self.K, max_iters=self.max_iters // 3, init='++')
kmeans.fit(self.X)
self.assignments = kmeans.predict()
self.means = kmeans.centroids
self.covs = []
for i in np.unique(self.assignments):
self.weights[int(i)] = (self.assignments == i).sum()
self.covs.append(np.cov(self.X[self.assignments == i].T))
else:
raise ValueError('Unknown type of init parameter')
self.weights /= self.weights.sum()
def KMeans_and_GMM(K):
COLOR = 'bgrcmyk'
X, y = make_clusters(skew=True, n_samples=1500, centers=K)
_, axes = plt.subplots(1, 3)
# Ground Truth
axes[0].scatter(X[:, 0], X[:, 1], c=[COLOR[int(assignment)] for assignment in y])
axes[0].set_title("Ground Truth")
# KMeans
kmeans = KMeans(K=K, init='++')
kmeans.fit(X)
y_kmeans = kmeans.predict()
c_kmeans = np.array(kmeans.centroids)
axes[1].scatter(X[:, 0], X[:, 1], c=[COLOR[int(assignment)] for assignment in y_kmeans])
axes[1].scatter(c_kmeans[:, 0], c_kmeans[:, 1], c=COLOR[:K], marker="o", s=500)
axes[1].set_title("KMeans")
# Gaussian Mixture
gmm = GaussianMixture(K=K, init='kmeans')
gmm.fit(X)
axes[2].set_title("Gaussian Mixture")
gmm.plot(ax=axes[2])
def iris_example(plot=False):
X, y = load_iris()
clusters = len(np.unique(y))
k = KMeans(K=clusters, max_iters=50, init="++")
k.fit(X, y)
k.predict()
data = np.zeros([k.n_samples, 2])
# Dimension reducing
# Sepal width*length
data[:, 0] = k.X[:, 0] * k.X[:, 1]
# Petal width*length
data[:, 1] = k.X[:, 2] * k.X[:, 3]
if plot:
k.plot(data)
def KMeans_and_GMM(K):
COLOR = 'bgrcmyk'
X, y = make_clusters(skew=True, n_samples=1500, centers=K)
_, axes = plt.subplots(1, 3)
# Ground Truth
axes[0].scatter(X[:, 0], X[:, 1], c=[COLOR[int(assignment)] for assignment in y])
axes[0].set_title("Ground Truth")
# KMeans
kmeans = KMeans(K=K, init='++')
kmeans.fit(X)
kmeans.predict()
axes[1].set_title("KMeans")
kmeans.plot(ax=axes[1], holdon=True)
# Gaussian Mixture
gmm = GaussianMixture(K=K, init='kmeans')
gmm.fit(X)
axes[2].set_title("Gaussian Mixture")
gmm.plot(ax=axes[2])
def KMeans_and_GMM(K):
COLOR = 'bgrcmyk'
X, y = make_clusters(skew=True, n_samples=1500, centers=K)
_, axes = plt.subplots(1, 3)
# Ground Truth
axes[0].scatter(X[:, 0], X[:, 1], c=[COLOR[int(assignment)] for assignment in y])
axes[0].set_title("Ground Truth")
# KMeans
kmeans = KMeans(K=K, init='++')
kmeans.fit(X)
kmeans.predict()
axes[1].set_title("KMeans")
kmeans.plot(ax=axes[1], holdon=True)
# Gaussian Mixture
gmm = GaussianMixture(K=K, init='kmeans')
gmm.fit(X)
axes[2].set_title("Gaussian Mixture")
gmm.plot(ax=axes[2])
def test_initialization():
with pytest.raises(ValueError):
kmeans = KMeans(init='test', K=2)
kmeans.fit(data)
kmeans._initialize_cetroids('test')
kmeans = KMeans(init='random', K=2)
kmeans.fit(data)
kmeans._initialize_cetroids('random')
assert len(kmeans.centroids) == kmeans.K
kmeans = KMeans(init='++', K=2)
kmeans.fit(data)
kmeans._initialize_cetroids('++')
assert len(kmeans.centroids) == kmeans.K