def test_err():
try:
km = knor.Kmeans(7.2)
return km.fit(FN, nrow=50, ncol=5)
except Exception as msg:
print(
("Successful test: CORRECTLY fails with message: {}".format(msg)))
def cluster(data, init0='random', clf='kmeans'):
if clf=='kmeans' or clf=='gmm':
if clf=='kmeans':
classifier = KMeans(n_clusters=2, init=init0, max_iter=1)
if clf=='kmeans-conv':
classifier = KMeans(n_clusters=2, init=init0)
elif clf=='gmm':
classifier = GaussianMixture(n_components=2, init_params=init0, covariance_type='diag')
t0 = time.time()
classifier.fit(data)
t1 = time.time()
dur = t1-t0
klabels = classifier.predict(data)
elif clf=='knor':
t0 = time.time()
ret = knor.Kmeans(data, 2)
t1 = time.time()
dur = t1-t0
klabels = ret.get_clusters()
else:
print('specify classifier')
if np.sum(klabels[:50]) > 25:
klabels = 1-klabels
klabels_0 = data[klabels==0]
klabels_1 = data[klabels==1]
half = len(data)//2
dlabels = np.concatenate(([0]*half,[1]*(len(data)-half)),axis=0)
err = np.sum(((dlabels!=klabels)))/len(data)
return klabels, klabels_0, klabels_1, err, dur
def heatmap(self, mode=None):
"""
Generate heatmap of all observations
If n > 1000, then a kmeans++ initializaiton is performed to derive
1000 cluster centers.
"""
if not mode:
mode = self._mode
if self._ds.n > 1000: #if sample size is > 1000, run kmeans++ initialization
ret = knor.Kmeans(self._ds_normed.D.values,
1000,
max_iters=0,
init='kmeanspp')
centroids_df = pd.DataFrame(ret.get_centroids(),
columns=self._ds.D.columns)
centroids_ds = lds.DataSet(centroids_df, name=self._ds.name)
return lpl.Heatmap(centroids_ds,
mode=mode).plot(showticklabels=True)
else:
return lpl.Heatmap(self._ds_normed,
mode=mode).plot(showticklabels=True)
def histogram_heatmap(self, mode=None):
"""
Generate 1d heatmap
"""
if not mode:
mode = self._mode
if self._ds.n > 1000: #if sample size is > 1000, run kmeans++ initialization
ret = knor.Kmeans(self._ds_normed.D.values,
1000,
max_iters=0,
init='kmeanspp')
centroids_df = pd.DataFrame(ret.get_centroids(),
columns=self._ds.D.columns)
centroids_ds = lds.DataSet(centroids_df, name=self._ds.name)
return lpl.HistogramHeatmap(
centroids_ds,
mode=mode).plot(showticklabels=self._showticklabels)
else:
return lpl.HistogramHeatmap(
self._ds_normed,
mode=mode).plot(showticklabels=self._showticklabels)
def dexm_test_c_im():
k = 2
centers = np.random.random((k, 5))
km = knor.Kmeans(k, centers=centers)
return km.fit(FN, 50, 5)
def dexm_test_c_comp():
km = knor.Kmeans(8)
return km.fit(FN, 50, 5)
def dim_test_c_im():
data = np.random.random((10, 3))
centers = np.random.random((3, 3))
km = knor.Kmeans(3, centers=centers) # TODO: Infer k
return km.fit(data)
def dim_test_c_comp():
data = np.random.random((10, 3))
km = knor.Kmeans(4)
return km.fit(data)
def dexm_test_c_im():
centers = np.random.random((2,5))
return knor.Kmeans(FN, centers, nrow=50, ncol=5)
def dexm_test_c_comp():
centers = 8
return knor.Kmeans(FN, centers, nrow=50, ncol=5)
def dim_test_c_im():
data = np.random.random((10,3))
centers = np.random.random((3,3))
return knor.Kmeans(data, centers)
def dim_test_c_comp():
data = np.random.random((10,3))
return knor.Kmeans(data, 4)