def plot_combined(p, x=None, y=None, indexes=None, k=iso_levels.DEFAULT_K):
"""Plot both, contour level stats and the actual contour plots for different values of k"""
p, x, y = _validate_infer_indexes(p, x, y, indexes)
# generate levels
try:
pdf_lvls = []
prob_lvls = []
prob_hori_lvls = []
for ki in k:
pdf_lvls.append(iso_levels.equi_value(p, ki))
prob_lvls.append(iso_levels.equi_prob_per_level(p, ki))
prob_hori_lvls.append(
iso_levels.equi_horizontal_prob_per_level(p, ki))
except TypeError:
pdf_lvls = [iso_levels.equi_value(p, k)]
prob_lvls = [iso_levels.equi_prob_per_level(p, k)]
prob_hori_lvls = [iso_levels.equi_horizontal_prob_per_level(p, k)]
k = [k]
# plot
for i, ki in enumerate(k):
levels = [pdf_lvls[i], prob_lvls[i], prob_hori_lvls[i]]
plot_contour_levels_stats(
levels,
len(levels) * [p],
labels=[
s.format(ki)
for s in ['old (k={})', 'vertical (k={})', 'horizonal (k={})']
])
for i, ki in enumerate(k):
contour_comparision_plot_2d(
[pdf_lvls[i], prob_lvls[i], prob_hori_lvls[i]],
p,
x,
y,
labels=[
s.format(ki)
for s in ['old (k={})', 'vertical (k={})', 'horizonal (k={})']
])
def plateau(k=5):
factor = 2.3
mu = np.array([-0.5, 1, 3.5, 7, 6]) - 3
sigma = np.array([0.05, 0.5, 0.2, 2, 0.3]) * 1.2
weights = np.array([0.2, 1, 0.4, 1, 3])
indexes_2d = utils.normalize_to_indexes(low=[-10, -10],
high=[10, 10],
n=100)
# mu = [0]
# sigma = [1]
# weights = [1]
# indexes_2d = utils.normalize_to_indexes(low=[-4, -4], high=[4, 4], n=100)
p = support_mixed_gaussian_2d(indexes_2d, mu, sigma, weights)
# raise p value
p_raised = p + np.max(p) * factor
# apply a circular 0--1 function
for ix, x in enumerate(indexes_2d[0]):
for iy, y in enumerate(indexes_2d[1]):
if x * x + y * y > 32:
p_raised[ix][iy] = 0
# plot
#plotting.plot_combined(p_raised, indexes=indexes_2d, k=[3,7,10])
levels = iso_levels.equi_prob_per_level(p_raised, k=k)
levels2 = iso_levels.equi_value(p_raised, k=k)
slice_ = utils.get_slice(p_raised, indexes_2d, 'y', 0)
fig, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_2d(p_raised,
levels2,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[0])
plotting.combined_2d(p_raised,
levels,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[1])
fig.show()
return fig
def data_file_with_kde(filepath,
kernel_bandwidth=None,
k=7,
usecols=None,
index_col=None):
# get data
df = pd.read_csv(filepath, index_col=index_col, usecols=usecols)
data = df.values.transpose()
# derive pdf
mykernel = pdf_kernel(data, kernel_bandwidth=kernel_bandwidth)
# get support
indexes = utils.normalize_to_indexes(data=data)
p = support_kde(mykernel, indexes)
# plot
#plotting.plot_combined(p, indexes[0], indexes[1], k=[3, 5, 7, 10])
levels = iso_levels.equi_prob_per_level(p, k=k)
levels2 = iso_levels.equi_value(p, k=k)
# get index of max
max_idx = np.unravel_index(np.argmax(p, axis=None), p.shape)
slice_ = utils.get_slice(p, indexes, 'y', indexes[1][max_idx[0]])
#slice_ = utils.get_slice(p, indexes, 'y', 68)
# print('old embrace ratio: {}'.format(stats.embrace_ratio(levels2, p)))
# print('new embrace ratio: {}'.format(stats.embrace_ratio(levels, p)))
fig, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_2d(p,
levels2,
x=indexes[0],
y=indexes[1],
slice_=slice_,
ax=ax[0])
plotting.combined_2d(p,
levels,
x=indexes[0],
y=indexes[1],
slice_=slice_,
ax=ax[1])
fig.show()
return fig
def broad_and_normal_gaussians(k=5):
# works well!
# mu = np.array([0, -5.5, 0, 5.5])
# sigma = np.array([1, 10, 6, 10])
# weights = np.array([1, 1, 0.5, 1])
# more complex and still works well
mu = np.array([[0, 0], [-5.5, -1], [0, 0], [5.5, 2], [-3, 4]])
sigma = np.array([1, 10, 6, 10, 9])
sigma = [[[s, 0], [0, s]] for s in sigma]
weights = np.array([1, 1, 0.5, 1, 1])
indexes_2d = utils.normalize_to_indexes(low=[-12, -10],
high=[12, 10],
n=100)
p = support_mixed_gaussian_2d(indexes_2d,
mu,
sigma,
weights,
from_scalar=False)
levels = iso_levels.equi_prob_per_level(p, k=k)
levels2 = iso_levels.equi_value(p, k=k)
slice_ = utils.get_slice(p, indexes_2d, 'y', 0)
fig, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_2d(p,
levels2,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[0])
plotting.combined_2d(p,
levels,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[1])
fig.show()
return fig
def iris_kde(kernel_bandwidth=None, k=6):
# get data
from sklearn import datasets
iris = datasets.load_iris()
data = iris.data[:, :2].transpose()
# derive pdf
mykernel = pdf_kernel(data, kernel_bandwidth=kernel_bandwidth)
# get support
indexes = utils.normalize_to_indexes(data=data)
p = support_kde(mykernel, indexes)
# plot
#plotting.plot_combined(p, indexes[0], indexes[1], k=[3, 5, 10])
levels = iso_levels.equi_prob_per_level(p, k=k)
levels2 = iso_levels.equi_value(p, k=k)
slice_ = utils.get_slice(p, indexes, 'y', 3)
#slice_ = None
fig, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_2d(p,
levels2,
x=indexes[0],
y=indexes[1],
slice_=slice_,
ax=ax[0])
plotting.combined_2d(p,
levels,
x=indexes[0],
y=indexes[1],
slice_=slice_,
ax=ax[1])
fig.show()
return fig
def basic_idea():
"""Creates plot for initial explanatory and motivating example for paper.
Also provide some search capabilities, i.e. allows to play with parameters to find exemplary distributions.
"""
mu = [0, 1, 3.5, 7]
sigma = [0.05, 0.5, 0.7, 2]
weights = [0.6, 1, 0.7, 1]
index_1d = utils.normalize_to_indexes(low=[-1], high=[10], n=2500, d=1)[0]
p_single_1d = [
support_gaussian_1d(index_1d, m, s) for m, s in zip(mu, sigma)
]
p_mixture_1d = sum(map(operator.mul, weights, p_single_1d))
levels = iso_levels.equi_prob_per_level(p_mixture_1d, k=7)
levels2 = iso_levels.equi_value(p_mixture_1d, k=7)
# I used this i identify suitable mu, sigma and weights
# figure = plt.figure(figsize=(9, 4))
# ax = figure.add_subplot(121)
# plotting.density(levels, gp1d, ax=ax)
# ax = figure.add_subplot(122)
# plotting.density(levels2, gp1d, ax=ax)
fig1d, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_1d(p_mixture_1d, levels2, index_1d, ax[0])
plotting.combined_1d(p_mixture_1d, levels, index_1d, ax[1])
fig1d.show()
indexes_2d = utils.normalize_to_indexes(low=[-1, -3], high=[10, 3], n=100)
p_mixture_2d = support_mixed_gaussian_2d(indexes_2d, mu, sigma, weights)
levels = iso_levels.equi_prob_per_level(p_mixture_2d, k=7)
levels2 = iso_levels.equi_value(p_mixture_2d, k=7)
slice_idx = int(len(indexes_2d[1]) / 2)
slice_val = indexes_2d[1][slice_idx]
slice_ = utils.get_slice(p_mixture_2d, indexes_2d, 'y', slice_val)
print('old embrace ratio: {}'.format(
stats.embrace_ratio(levels2, p_mixture_2d)))
print('new embrace ratio: {}'.format(
stats.embrace_ratio(levels, p_mixture_2d)))
# I used this to identify k=7 as particularly interesting
#plotting.plot_combined(gp2d, indexes=gindex, k=list(range(2,10)))
#plotting.plot_combined(gp2d, indexes=gindex, k=[7])
fig2d, ax = plt.subplots(2, 3, figsize=(3 * 5, 8))
plotting.combined_2d(p_mixture_2d,
levels2,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[0])
plotting.combined_2d(p_mixture_2d,
levels,
x=indexes_2d[0],
y=indexes_2d[1],
slice_=slice_,
ax=ax[1])
fig2d.show()
return fig1d, fig2d