from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
dataset = get_pendigits().rescale()
# X = dataset.X
pca = PCA(2)
pca.fit(dataset.X)
pca_X = pca.transform(dataset.X)
# print('X:', X)
bandwidths = {
'gaussian_dist': BandwidthSelection.gaussian_distribution(pca_X),
'likelihood_10': BandwidthSelection.cv_maximum_likelihood(pca_X),
'likelihood_100': BandwidthSelection.cv_maximum_likelihood(pca_X, search=np.linspace(1e-4, 1, 100)),
}
# Set up the figure
f, axes = plt.subplots(int(len(bandwidths / 4)) + 1, 4, figsize=(8, 8))
x, y = list(map(np.array, zip(*pca_X)))
bandwidths = sorted([(k, v) for k, v in bandwidths.items()], key=lambda x: x[0])
for ax, (name, bandwidth) in zip(axes.flat, bandwidths):
ax.set_aspect("equal")
# Draw the two density plots
sns.kdeplot(x, y, bw=bandwidth,
cmap="Reds", shade=True, shade_lowest=False, ax=ax)
from dataset import *
from bandwidth_selection import BandwidthSelection
datasets = [
get_iris_with_test(bandwidth='cv_ml').rescale(),
get_pendigits(bandwidth='cv_ml').rescale(),
get_yeast_with_test(bandwidth='cv_ml').rescale(),
get_satimage(bandwidth='cv_ml').rescale(),
get_banknote_with_test(bandwidth='cv_ml').rescale(),
get_spam_with_test(bandwidth='cv_ml').rescale(),
get_drd_with_test(bandwidth='cv_ml').rescale(),
get_imagesegment(bandwidth='cv_ml').rescale(),
get_pageblock_with_test(bandwidth='cv_ml').rescale(),
get_statlogsegment_with_test(bandwidth='cv_ml').rescale(),
get_winequality_with_test('white', bandwidth='cv_ml').rescale(),
get_winequality_with_test('red', bandwidth='cv_ml').rescale(),
]
datasets.sort(key=lambda dataset: len(dataset.X))
for dataset in datasets:
print('generating for', dataset.name)
bw = BandwidthSelection.gaussian_distribution(dataset.X)
print('bw:', 'rot:', bw, 'cv_ml:', dataset.bandwidth)