def mi(x, y, bins=10):
"""Mutual information between x and y"""
H_x = u.compute_entropy(np.histogram(x, bins)[0])
H_y = u.compute_entropy(np.histogram(y, bins)[0])
c_xy = np.histogram2d(x, y, bins)[0]
mi = skm.mutual_info_score(None, None, contingency=c_xy)
return mi / np.sqrt(H_x * H_y)
def mi(x, y, bins=10):
"""Mutual information between x and y"""
H_x = u.compute_entropy(np.histogram(x, bins)[0])
H_y = u.compute_entropy(np.histogram(y, bins)[0])
c_xy = np.histogram2d(x, y, bins)[0]
mi = skm.mutual_info_score(None, None, contingency=c_xy)
return mi / np.sqrt(H_x * H_y)
def get_branch_lengths_estimates(tree): """ :param tree: Tree node or tree file or newick tree string; :return: """ # TBL branches = get_branch_lengths(tree) entropy = compute_entropy(branches) return max(branches), min(branches), np.mean(branches), np.std(branches), entropy
def get_diameters_estimates(tree_filepath, actual_bl=True): """ if not actual_bl - function changes the tree! send only filepath :param tree_filepath: tree file or newick tree string; :param actual_bl: True to sum actual dists, False for num of branches :return: min, max, mean, and std of tree diameters """ # tree = copy.deepcopy(get_newick_tree(tree)) # do not deepcopy! when trees are large it exceeds recursion depth if not actual_bl: assert isinstance(tree_filepath, str) tree = get_newick_tree(tree_filepath) tree_root = tree.get_tree_root() if not actual_bl: for node in tree_root.iter_descendants(): node.dist = 1.0 tree_diams = [] leaves = list(tree_root.iter_leaves()) for leaf1, leaf2 in itertools.combinations(leaves, 2): tree_diams.append(leaf1.get_distance(leaf2)) entropy = compute_entropy(tree_diams) return max(tree_diams), min(tree_diams), np.mean(tree_diams), np.std(tree_diams), entropy
def time_entropy(visits):
"""Compute entropy of venue with respect to time of the day of its
checkins."""
hours = np.bincount([t.hour for t in visits], minlength=24)
return u.compute_entropy(hours.astype(float))/np.log(24.0)
def venue_entropy(visitors):
"""Compute the entropy of venue given the list of its `visitors`."""
# pylint: disable=E1101
return u.compute_entropy(np.array(Counter(visitors).values(), dtype=float))
def jensen_shannon_divergence(P, Q):
"""Compute JSD(P || Q) as defined in
https://en.wikipedia.org/wiki/Jensen–Shannon_divergence """
avg = 0.5*(P + Q)
avg_entropy = 0.5*(u.compute_entropy(P) + u.compute_entropy(Q))
return u.compute_entropy(avg) - avg_entropy
def jensen_shannon_divergence(P, Q):
"""Compute JSD(P || Q) as defined in
https://en.wikipedia.org/wiki/Jensen–Shannon_divergence """
avg = 0.5 * (P + Q)
avg_entropy = 0.5 * (u.compute_entropy(P) + u.compute_entropy(Q))
return u.compute_entropy(avg) - avg_entropy
def make_bias_noise_figure(bias_cp_dfs,
entropy_cp_dfs,
noisy_cp_dfs,
noise_levels,
noiseless_cp_df,
analytical_cp_df,
bias_cp_df_labels=None,
entropy_cp_df_labels=None,
base_figure_scale=3,
include_binomial_null=False):
num_noisy_cp_dfs = len(noisy_cp_dfs)
normal_axis_scale = 5
small_axis_scale = 2
gap_scale = 1
figsize = (num_noisy_cp_dfs * base_figure_scale, 2 * base_figure_scale)
gridsize = [
normal_axis_scale * 2 + small_axis_scale + gap_scale,
num_noisy_cp_dfs * normal_axis_scale +
(num_noisy_cp_dfs - 1) * gap_scale
]
f, ax_grid = plt.subplots(*gridsize, figsize=figsize)
bias_ax = plt.subplot2grid(gridsize, (0, 0),
colspan=gridsize[-1] - normal_axis_scale -
gap_scale,
rowspan=normal_axis_scale)
counts = [
bias_cp_df.groupby("index_set").count().final_loss
for bias_cp_df in bias_cp_dfs
]
marginal_counts = [
utils.compute_marginal_counts(count) for count in counts
]
panels.make_bias_panel(bias_cp_dfs,
counts,
marginal_counts,
labels=bias_cp_df_labels,
ax=bias_ax,
include_binomial_null=include_binomial_null)
entropy_ax = plt.subplot2grid(gridsize,
(0, gridsize[-1] - normal_axis_scale),
rowspan=normal_axis_scale,
colspan=normal_axis_scale)
entropies = [utils.compute_entropy(cp_df) for cp_df in entropy_cp_dfs]
entropy_sds = [
utils.bootstrap_entropy_sd(cp_df) for cp_df in entropy_cp_dfs
]
panels.make_entropy_panel(entropy_cp_dfs,
entropies,
entropy_sds,
entropy_cp_df_labels,
ax=entropy_ax)
noise_axs = []
for ii, (noisy_cp_df,
noise_level) in enumerate(zip(noisy_cp_dfs, noise_levels)):
if not noise_axs == []:
subplot2grid_kwargs = {
"sharex": noise_axs[0],
"sharey": noise_axs[0]
}
else:
subplot2grid_kwargs = {}
noise_ax = plt.subplot2grid(gridsize,
(normal_axis_scale + gap_scale,
(normal_axis_scale + gap_scale) * ii),
rowspan=normal_axis_scale,
colspan=normal_axis_scale,
**subplot2grid_kwargs)
include_x_label = False
if ii == 0:
include_y_label = True
else:
include_y_label = False
if ii + 1 == num_noisy_cp_dfs:
include_legend = True
else:
include_legend = False
panels.make_noise_comparison_panel(noisy_cp_df,
analytical_cp_df,
noiseless_cp_df,
ax=noise_ax,
include_legend=include_legend,
include_x_label=include_x_label,
include_y_label=include_y_label,
noise_level=noise_level)
if subplot2grid_kwargs != {}:
subplot2grid_kwargs = {"sharex": noise_axs[0]}
histogram_ax = plt.subplot2grid(gridsize,
(normal_axis_scale * 2 + gap_scale,
(normal_axis_scale + gap_scale) * ii),
rowspan=small_axis_scale,
colspan=normal_axis_scale,
**subplot2grid_kwargs)
cps = [noisy_cp_df, analytical_cp_df]
colors = [NUMERICAL_CPS_COLOR, ANALYTICAL_CPS_COLOR]
panels.make_histogram_comparison_panel(cps,
colors,
ax=histogram_ax,
include_x_label=True,
include_y_label=include_y_label)
noise_axs.append(noise_ax)
return f, ax_grid, bias_ax, noise_axs
