def parse_distribution(distribution, model_type=None):
distribution = normalise_string(distribution)
if model_type is None:
kind = "reconstruction"
distributions = DISTRIBUTIONS
elif isinstance(model_type, str):
kind = "latent"
if model_type == "VAE":
distributions = LATENT_DISTRIBUTIONS
elif model_type == "GMVAE":
distributions = GAUSSIAN_MIXTURE_DISTRIBUTIONS
else:
raise ValueError("Model type not found.")
else:
raise TypeError("`model_type` should be a string.")
distribution_names = list(distributions.keys())
parsed_distribution_name = None
for distribution_name in distribution_names:
if normalise_string(distribution_name) == distribution:
parsed_distribution_name = distribution_name
if parsed_distribution_name is None:
raise ValueError("{} distribution `{}` not supported{}.".format(
kind.capitalize(), distribution,
" for {}".format(model_type) if model_type else ""))
return parsed_distribution_name
def _build_preprocessed_path(self,
map_features=None,
preprocessing_methods=None,
feature_selection=None,
feature_selection_parameters=None,
example_filter=None,
example_filter_parameters=None,
splitting_method=None,
splitting_fraction=None,
split_indices=None):
base_path = os.path.join(self.preprocess_directory, self.name)
filename_parts = [base_path]
if map_features:
filename_parts.append("features_mapped")
if feature_selection:
feature_selection_part = normalise_string(feature_selection)
if feature_selection_parameters:
for parameter in feature_selection_parameters:
feature_selection_part += "_" + normalise_string(
str(parameter))
filename_parts.append(feature_selection_part)
if example_filter:
example_filter_part = normalise_string(example_filter)
if example_filter_parameters:
for parameter in example_filter_parameters:
example_filter_part += "_" + normalise_string(
str(parameter))
filename_parts.append(example_filter_part)
if preprocessing_methods:
filename_parts.extend(map(normalise_string, preprocessing_methods))
if splitting_method:
filename_parts.append("split")
if (splitting_method == "indices" and len(split_indices) == 3
or not splitting_fraction):
filename_parts.append(splitting_method)
else:
filename_parts.append("{}_{}".format(splitting_method,
splitting_fraction))
path = "-".join(filename_parts) + PREPROCESSED_EXTENSION
return path
def parse_model_versions(proposed_versions):
version_alias_sets = {
"end_of_training": ["eot", "end", "finish", "finished"],
"best_model": ["bm", "best", "optimal_parameters", "op", "optimal"],
"early_stopping": ["es", "early", "stop", "stopped"]
}
parsed_versions = []
if not isinstance(proposed_versions, list):
proposed_versions = [proposed_versions]
if proposed_versions == ["all"]:
parsed_versions = list(version_alias_sets.keys())
else:
for proposed_version in proposed_versions:
normalised_proposed_version = normalise_string(proposed_version)
parsed_version = None
for version, version_aliases in version_alias_sets.items():
if (normalised_proposed_version == version
or normalised_proposed_version in version_aliases):
parsed_version = version
break
if parsed_version:
parsed_versions.append(parsed_version)
else:
raise ValueError(
"`{}` is not a model version.".format(proposed_version))
return parsed_versions
def save(data_dictionary, tables_file, group_title=None):
if group_title:
group = tables_file.create_group("/",
normalise_string(group_title),
group_title)
else:
group = tables_file.root
for title, value in data_dictionary.items():
if isinstance(value, scipy.sparse.csr_matrix):
_save_sparse_matrix(value, title, group, tables_file)
elif isinstance(value, (numpy.ndarray, list)):
_save_array(value, title, group, tables_file)
elif title == "split indices":
_save_split_indices(value, title, group, tables_file)
elif title == "feature mapping":
_save_feature_mapping(value, title, group, tables_file)
elif value is None:
_save_string(str(value), title, group, tables_file)
elif title.endswith("set"):
save(value, tables_file, group_title=title)
else:
raise NotImplementedError(
"Saving type {} for title \"{}\" has not been implemented."
.format(type(value), title))
def decorator(function):
aliases = set()
alias = normalise_string(name)
aliases.add(alias)
alias = alias.replace("_", "")
aliases.add(alias)
PREDICTION_METHODS[name] = {"aliases": aliases, "function": function}
return function
def name(self):
name_parts = [self.method, self.number_of_clusters]
if self.training_set_kind and self.training_set_kind != "training":
name_parts.append(self.training_set_kind)
name = "_".join(
map(lambda s: normalise_string(str(s)).replace("_", ""),
name_parts))
return name
def _save_sparse_matrix(sparse_matrix, title, group, tables_file):
name = normalise_string(title)
group = tables_file.create_group(group, name, title)
for attribute in ("data", "indices", "indptr", "shape"):
array = numpy.array(getattr(sparse_matrix, attribute))
_save_array(array, attribute, group, tables_file)
def _save_split_indices(split_indices, title, group, tables_file):
name = normalise_string(title)
group = tables_file.create_group(group, name, title)
for subset_name, subset_slice in split_indices.items():
subset_slice_array = numpy.array(
[subset_slice.start, subset_slice.stop])
_save_array(subset_slice_array, subset_name, group, tables_file)
def plot_centroid_covariance_matrices_evolution(covariance_matrices,
distribution,
name=None):
distribution = normalise_string(distribution)
figure_name = "centroids_evolution-{}-covariance_matrices".format(
distribution)
figure_name = saving.build_figure_name(figure_name, name)
y_label = _axis_label_for_symbol(symbol="\\Sigma",
distribution=distribution,
prefix="|",
suffix="(y = k)|")
n_epochs, n_centroids, __, __ = covariance_matrices.shape
determinants = numpy.empty([n_epochs, n_centroids])
for e in range(n_epochs):
for k in range(n_centroids):
determinants[e,
k] = numpy.prod(numpy.diag(covariance_matrices[e, k]))
if determinants.all() > 0:
line_range_ratio = numpy.empty(n_centroids)
for k in range(n_centroids):
determinants_min = determinants[:, k].min()
determinants_max = determinants[:, k].max()
line_range_ratio[k] = determinants_max / determinants_min
range_ratio = line_range_ratio.max() / line_range_ratio.min()
if range_ratio > 1e2:
y_scale = "log"
else:
y_scale = "linear"
centroids_palette = style.darker_palette(n_centroids)
epochs = numpy.arange(n_epochs) + 1
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.despine()
for k in range(n_centroids):
axis.plot(epochs,
determinants[:, k],
color=centroids_palette[k],
label="$k = {}$".format(k))
axis.set_xlabel("Epochs")
axis.set_ylabel(y_label)
axis.set_yscale(y_scale)
axis.legend(loc="best")
return figure, figure_name
def _axis_label_for_symbol(symbol,
coordinate=None,
decomposition_method=None,
distribution=None,
prefix="",
suffix=""):
if decomposition_method:
decomposition_method = proper_string(
normalise_string(decomposition_method), DECOMPOSITION_METHOD_NAMES)
decomposition_label = DECOMPOSITION_METHOD_LABEL[decomposition_method]
else:
decomposition_label = ""
if decomposition_label:
decomposition_label = "\\mathrm{{{}}}".format(decomposition_label)
if coordinate:
coordinate_text = "{{{} {}}}".format(decomposition_label, coordinate)
else:
coordinate_text = ""
if distribution == "prior":
distribution_symbol = "\\theta"
elif distribution == "posterior":
distribution_symbol = "\\phi"
else:
distribution_symbol = ""
if distribution_symbol and coordinate_text:
distribution_position = "_"
coordinate_position = "^"
elif distribution_symbol and not coordinate_text:
distribution_position = "_"
coordinate_position = ""
elif not distribution_symbol and coordinate_text:
distribution_position = ""
coordinate_position = "_"
else:
distribution_position = ""
coordinate_position = ""
if coordinate_position == "^":
coordinate_text = "{{(" + coordinate_text + ")}}"
elif coordinate_position == "_":
coordinate_text = "{{" + coordinate_text + "}}"
axis_label = "$" + "".join([
prefix, symbol, distribution_position, distribution_symbol,
coordinate_position, coordinate_text, suffix
]) + "$"
return axis_label
def _save_array(array, title, group, tables_file):
name = normalise_string(title)
if isinstance(array, list):
array = numpy.array(array)
name += "_was_list"
if array.dtype.char == "U":
encode = numpy.vectorize(lambda s: s.encode("UTF-8"))
array = encode(array).astype("S")
atom = tables.Atom.from_dtype(array.dtype)
data_store = tables_file.create_carray(group, name, atom, array.shape,
title)
data_store[:] = array
def _find_list_of_names(list_name_guesses, kind):
if list_name_guesses is None:
list_name_guesses = LIST_NAME_GUESSES[kind]
elif not isinstance(list_name_guesses, list):
list_name_guesses = [list_name_guesses]
list_of_names = None
for list_name_guess in list_name_guesses:
for table_key in table:
if list_name_guess == normalise_string(table_key):
list_of_names = table[table_key]
if list_of_names is not None:
break
return list_of_names
def find_data_set(name, directory):
data_sets = _load_data_set_metadata()
title = None
data_set = None
json_path = os.path.join(directory, name, name + ".json")
if os.path.exists(json_path):
title, data_set = _data_set_from_json_file(json_path)
if not title:
for data_set_title, data_set_specifications in data_sets.items():
if normalise_string(data_set_title) == normalise_string(name):
title = data_set_title
data_set = data_set_specifications
break
if not title:
raise KeyError("Data set not found.")
return title, data_set
def _find_list_of_names(list_name_guesses, kind):
if list_name_guesses is None:
list_name_guesses = LIST_NAME_GUESSES[kind]
elif not isinstance(list_name_guesses, list):
list_name_guesses = [list_name_guesses]
list_of_names = None
for list_name_guess in list_name_guesses:
for table_key in table:
if list_name_guess == normalise_string(table_key):
list_of_names = table[table_key]
if list_of_names is not None:
break
list_of_names = numpy.array(
["{} {}".format(kind, i + 1) for i in range(n[kind])])
def default_feature_parameters(self):
feature_selection_parameters = None
if self.feature_selection:
feature_selection = normalise_string(self.feature_selection)
if feature_selection == "keep_variances_above":
feature_selection_parameters = [0.5]
elif feature_selection == "keep_highest_variances":
if self.number_of_features is not None:
feature_selection_parameters = [
int(self.number_of_features / 2)
]
return feature_selection_parameters
def save_values(values,
name,
row_names=None,
column_names=None,
directory=None):
safe_name = "-".join([normalise_string(part) for part in name.split("-")])
filename = "{}.tsv.gz".format(safe_name)
path = os.path.join(directory, filename)
table = pandas.DataFrame(data=values,
index=row_names,
columns=column_names)
if not os.path.exists(directory):
os.makedirs(directory)
table.to_csv(path, sep="\t")
def build_figure_name(base_name, other_names=None):
if isinstance(base_name, list):
if not other_names:
other_names = []
other_names.extend(base_name[1:])
base_name = normalise_string(base_name[0])
figure_name = base_name
if other_names:
if not isinstance(other_names, list):
other_names = str(other_names)
other_names = [other_names]
else:
other_names = [
str(name) for name in other_names if name is not None
]
figure_name += "-" + "-".join(map(normalise_string, other_names))
return figure_name
def parse_input(input_file_or_name):
if input_file_or_name.endswith(".json"):
json_path = input_file_or_name
with open(json_path, "r") as json_file:
data_set_dictionary = json.load(json_file)
name = _base_name(json_path)
if "URLs" not in data_set_dictionary:
if "values" in data_set_dictionary:
json_directory = os.path.dirname(json_path)
data_set_dictionary["values"] = os.path.join(
json_directory, data_set_dictionary["values"])
else:
raise KeyError("Missing path or URL to values.")
if "labels" in data_set_dictionary:
json_directory = os.path.dirname(json_path)
data_set_dictionary["labels"] = os.path.join(
json_directory, data_set_dictionary["labels"])
elif os.path.isfile(input_file_or_name):
file_path = input_file_or_name
filename = os.path.basename(file_path)
file_extension = extension(filename)
data_format = file_extension[1:] if file_extension else None
name = _base_name(file_path)
data_set_dictionary = {"values": file_path, "format": data_format}
else:
name = input_file_or_name
name = normalise_string(name)
data_set_dictionary = None
return name, data_set_dictionary
def _save_feature_mapping(feature_mapping, title, group, tables_file):
name = normalise_string(title)
group = tables_file.create_group(group, name, title)
feature_names = []
feature_counts = []
feature_ids = []
for feature_name, feature_id_set in feature_mapping.items():
feature_names.append(feature_name)
feature_counts.append(len(feature_id_set))
feature_ids.extend(feature_id_set)
feature_lists = {
"feature_names": feature_names,
"feature_counts": feature_counts,
"feature_ids": feature_ids
}
for feature_list_name, feature_list in feature_lists.items():
feature_list_array = numpy.array(feature_list)
_save_array(feature_list_array, feature_list_name, group, tables_file)
def __init__(self,
method,
number_of_clusters=None,
training_set_kind=None):
prediction_method_names = {
name: specifications["aliases"]
for name, specifications in PREDICTION_METHODS.items()
}
method = proper_string(method, prediction_method_names)
if method not in PREDICTION_METHODS:
raise ValueError(
"Prediction method `{}` not found.".format(method))
if number_of_clusters is None:
raise TypeError("Number of clusters not set.")
self.method = method
self.number_of_clusters = number_of_clusters
if training_set_kind:
training_set_kind = normalise_string(training_set_kind)
self.training_set_kind = training_set_kind
def plot_centroid_probabilities_evolution(probabilities,
distribution,
linestyle="solid",
name=None):
distribution = normalise_string(distribution)
y_label = _axis_label_for_symbol(symbol="\\pi",
distribution=distribution,
suffix="^k")
figure_name = "centroids_evolution-{}-probabilities".format(distribution)
figure_name = saving.build_figure_name(figure_name, name)
n_epochs, n_centroids = probabilities.shape
centroids_palette = style.darker_palette(n_centroids)
epochs = numpy.arange(n_epochs) + 1
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.despine()
for k in range(n_centroids):
axis.plot(epochs,
probabilities[:, k],
color=centroids_palette[k],
linestyle=linestyle,
label="$k = {}$".format(k))
axis.set_xlabel("Epochs")
axis.set_ylabel(y_label)
axis.legend(loc="best")
return figure, figure_name
def plot_centroid_means_evolution(means,
distribution,
decomposed=False,
name=None):
symbol = "\\mu"
if decomposed:
decomposition_method = "PCA"
else:
decomposition_method = ""
distribution = normalise_string(distribution)
suffix = "(y = k)"
x_label = _axis_label_for_symbol(symbol=symbol,
coordinate=1,
decomposition_method=decomposition_method,
distribution=distribution,
suffix=suffix)
y_label = _axis_label_for_symbol(symbol=symbol,
coordinate=2,
decomposition_method=decomposition_method,
distribution=distribution,
suffix=suffix)
figure_name = "centroids_evolution-{}-means".format(distribution)
figure_name = saving.build_figure_name(figure_name, name)
n_epochs, n_centroids, latent_size = means.shape
if latent_size > 2:
raise ValueError("Dimensions of means should be 2.")
centroids_palette = style.darker_palette(n_centroids)
epochs = numpy.arange(n_epochs) + 1
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.despine()
colour_bar_scatter_plot = axis.scatter(means[:, 0, 0],
means[:, 0, 1],
c=epochs,
cmap=seaborn.dark_palette(
style.NEUTRAL_COLOUR,
as_cmap=True),
zorder=0)
for k in range(n_centroids):
colour = centroids_palette[k]
colour_map = seaborn.dark_palette(colour, as_cmap=True)
axis.plot(means[:, k, 0],
means[:, k, 1],
color=colour,
label="$k = {}$".format(k),
zorder=k + 1)
axis.scatter(means[:, k, 0],
means[:, k, 1],
c=epochs,
cmap=colour_map,
zorder=n_centroids + k + 1)
axis.legend(loc="best")
colour_bar = figure.colorbar(colour_bar_scatter_plot)
colour_bar.outline.set_linewidth(0)
colour_bar.set_label("Epochs")
axis.set_xlabel(x_label)
axis.set_ylabel(y_label)
return figure, figure_name
def select_features(values_dictionary, feature_names, method=None,
parameters=None):
method = normalise_string(method)
print("Selecting features.")
start_time = time()
if type(values_dictionary) == dict:
values = values_dictionary["original"]
n_examples, n_features = values.shape
if method == "remove_zeros":
total_feature_sum = values.sum(axis=0)
if isinstance(total_feature_sum, numpy.matrix):
total_feature_sum = total_feature_sum.A.squeeze()
indices = total_feature_sum != 0
elif method == "keep_variances_above":
variances = values.var(axis=0)
if isinstance(variances, numpy.matrix):
variances = variances.A.squeeze()
if parameters:
threshold = float(parameters[0])
else:
threshold = 0.5
indices = variances > threshold
elif method == "keep_highest_variances":
variances = values.var(axis=0)
if isinstance(variances, numpy.matrix):
variances = variances.A.squeeze()
variance_sorted_indices = numpy.argsort(variances)
if parameters:
number_to_keep = int(parameters[0])
else:
number_to_keep = int(n_examples/2)
indices = numpy.sort(variance_sorted_indices[-number_to_keep:])
else:
raise ValueError(
"Feature selection `{}` not found.".format(method))
if method:
error = Exception(
"No features excluded using feature selection {}.".format(method))
if indices.dtype == "bool" and all(indices):
raise error
elif indices.dtype != "bool" and len(indices) == n_features:
raise error
feature_selected_values = {}
for version, values in values_dictionary.items():
if values is not None:
feature_selected_values[version] = values[:, indices]
else:
feature_selected_values[version] = None
feature_selected_feature_names = feature_names[indices]
n_features_changed = len(feature_selected_feature_names)
duration = time() - start_time
print("{} features selected, {} excluded ({}).".format(
n_features_changed,
n_features - n_features_changed,
format_duration(duration)
))
return feature_selected_values, feature_selected_feature_names
def plot_values(values,
colour_coding=None,
colouring_data_set=None,
centroids=None,
sampled_values=None,
class_name=None,
feature_index=None,
figure_labels=None,
example_tag=None,
name="scatter"):
figure_name = name
if figure_labels:
title = figure_labels.get("title")
x_label = figure_labels.get("x label")
y_label = figure_labels.get("y label")
else:
title = "none"
x_label = "$x$"
y_label = "$y$"
if not title:
title = "none"
figure_name += "-" + normalise_string(title)
if colour_coding:
colour_coding = normalise_string(colour_coding)
figure_name += "-" + colour_coding
if "predicted" in colour_coding:
if colouring_data_set.prediction_specifications:
figure_name += "-" + (
colouring_data_set.prediction_specifications.name)
else:
figure_name += "unknown_prediction_method"
if colouring_data_set is None:
raise ValueError("Colouring data set not given.")
if sampled_values is not None:
figure_name += "-samples"
values = values.copy()[:, :2]
if scipy.sparse.issparse(values):
values = values.A
# Randomise examples in values to remove any prior order
n_examples, __ = values.shape
random_state = numpy.random.RandomState(117)
shuffled_indices = random_state.permutation(n_examples)
values = values[shuffled_indices]
# Adjust marker size based on number of examples
style._adjust_marker_size_for_scatter_plots(n_examples)
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.despine()
axis.set_xlabel(x_label)
axis.set_ylabel(y_label)
colour_map = seaborn.dark_palette(style.STANDARD_PALETTE[0], as_cmap=True)
alpha = 1
if sampled_values is not None:
alpha = 0.5
if colour_coding and ("labels" in colour_coding or "ids" in colour_coding
or "class" in colour_coding
or colour_coding == "batches"):
if colour_coding == "predicted_cluster_ids":
labels = colouring_data_set.predicted_cluster_ids
class_names = numpy.unique(labels).tolist()
number_of_classes = len(class_names)
class_palette = None
label_sorter = None
elif colour_coding == "predicted_labels":
labels = colouring_data_set.predicted_labels
class_names = colouring_data_set.predicted_class_names
number_of_classes = colouring_data_set.number_of_predicted_classes
class_palette = colouring_data_set.predicted_class_palette
label_sorter = colouring_data_set.predicted_label_sorter
elif colour_coding == "predicted_superset_labels":
labels = colouring_data_set.predicted_superset_labels
class_names = colouring_data_set.predicted_superset_class_names
number_of_classes = (
colouring_data_set.number_of_predicted_superset_classes)
class_palette = colouring_data_set.predicted_superset_class_palette
label_sorter = colouring_data_set.predicted_superset_label_sorter
elif "superset" in colour_coding:
labels = colouring_data_set.superset_labels
class_names = colouring_data_set.superset_class_names
number_of_classes = colouring_data_set.number_of_superset_classes
class_palette = colouring_data_set.superset_class_palette
label_sorter = colouring_data_set.superset_label_sorter
elif colour_coding == "batches":
labels = colouring_data_set.batch_indices.flatten()
class_names = colouring_data_set.batch_names
number_of_classes = colouring_data_set.number_of_batches
class_palette = None
label_sorter = None
else:
labels = colouring_data_set.labels
class_names = colouring_data_set.class_names
number_of_classes = colouring_data_set.number_of_classes
class_palette = colouring_data_set.class_palette
label_sorter = colouring_data_set.label_sorter
if not class_palette:
index_palette = style.lighter_palette(number_of_classes)
class_palette = {
class_name: index_palette[i]
for i, class_name in enumerate(
sorted(class_names, key=label_sorter))
}
# Examples are shuffled, so should their labels be
labels = labels[shuffled_indices]
if ("labels" in colour_coding or "ids" in colour_coding
or colour_coding == "batches"):
colours = []
classes = set()
for i, label in enumerate(labels):
colour = class_palette[label]
colours.append(colour)
# Plot one example for each class to add labels
if label not in classes:
classes.add(label)
axis.scatter(values[i, 0],
values[i, 1],
color=colour,
label=label,
alpha=alpha)
axis.scatter(values[:, 0], values[:, 1], c=colours, alpha=alpha)
class_handles, class_labels = axis.get_legend_handles_labels()
if class_labels:
class_labels, class_handles = zip(
*sorted(zip(class_labels, class_handles),
key=(lambda t: label_sorter(t[0])
) if label_sorter else None))
class_label_maximum_width = max(map(len, class_labels))
if class_label_maximum_width <= 5 and number_of_classes <= 20:
axis.legend(class_handles, class_labels, loc="best")
else:
if number_of_classes <= 20:
class_label_columns = 2
else:
class_label_columns = 3
axis.legend(
class_handles,
class_labels,
bbox_to_anchor=(-0.1, 1.05, 1.1, 0.95),
loc="lower left",
ncol=class_label_columns,
mode="expand",
borderaxespad=0.,
)
elif "class" in colour_coding:
colours = []
figure_name += "-" + normalise_string(str(class_name))
ordered_indices_set = {str(class_name): [], "Remaining": []}
for i, label in enumerate(labels):
if label == class_name:
colour = class_palette[label]
ordered_indices_set[str(class_name)].append(i)
else:
colour = style.NEUTRAL_COLOUR
ordered_indices_set["Remaining"].append(i)
colours.append(colour)
colours = numpy.array(colours)
z_order_index = 1
for label, ordered_indices in sorted(ordered_indices_set.items()):
if label == "Remaining":
z_order = 0
else:
z_order = z_order_index
z_order_index += 1
ordered_values = values[ordered_indices]
ordered_colours = colours[ordered_indices]
axis.scatter(ordered_values[:, 0],
ordered_values[:, 1],
c=ordered_colours,
label=label,
alpha=alpha,
zorder=z_order)
handles, labels = axis.get_legend_handles_labels()
labels, handles = zip(*sorted(zip(labels, handles),
key=lambda t: label_sorter(t[0])
if label_sorter else None))
axis.legend(handles,
labels,
bbox_to_anchor=(-0.1, 1.05, 1.1, 0.95),
loc="lower left",
ncol=2,
mode="expand",
borderaxespad=0.)
elif colour_coding == "count_sum":
n = colouring_data_set.count_sum[shuffled_indices].flatten()
scatter_plot = axis.scatter(values[:, 0],
values[:, 1],
c=n,
cmap=colour_map,
alpha=alpha)
colour_bar = figure.colorbar(scatter_plot)
colour_bar.outline.set_linewidth(0)
colour_bar.set_label("Total number of {}s per {}".format(
colouring_data_set.terms["item"],
colouring_data_set.terms["example"]))
elif colour_coding == "feature":
if feature_index is None:
raise ValueError("Feature number not given.")
if feature_index > colouring_data_set.number_of_features:
raise ValueError("Feature number higher than number of features.")
feature_name = colouring_data_set.feature_names[feature_index]
figure_name += "-{}".format(normalise_string(feature_name))
f = colouring_data_set.values[shuffled_indices, feature_index]
if scipy.sparse.issparse(f):
f = f.A
f = f.squeeze()
scatter_plot = axis.scatter(values[:, 0],
values[:, 1],
c=f,
cmap=colour_map,
alpha=alpha)
colour_bar = figure.colorbar(scatter_plot)
colour_bar.outline.set_linewidth(0)
colour_bar.set_label(feature_name)
elif colour_coding is None:
axis.scatter(values[:, 0],
values[:, 1],
c="k",
alpha=alpha,
edgecolors="none")
else:
raise ValueError("Colour coding `{}` not found.".format(colour_coding))
if centroids:
prior_centroids = centroids["prior"]
if prior_centroids:
n_centroids = prior_centroids["probabilities"].shape[0]
else:
n_centroids = 0
if n_centroids > 1:
centroids_palette = style.darker_palette(n_centroids)
classes = numpy.arange(n_centroids)
means = prior_centroids["means"]
covariance_matrices = prior_centroids["covariance_matrices"]
for k in range(n_centroids):
axis.scatter(means[k, 0],
means[k, 1],
s=60,
marker="x",
color="black",
linewidth=3)
axis.scatter(means[k, 0],
means[k, 1],
marker="x",
facecolor=centroids_palette[k],
edgecolors="black")
ellipse_fill, ellipse_edge = _covariance_matrix_as_ellipse(
covariance_matrices[k],
means[k],
colour=centroids_palette[k])
axis.add_patch(ellipse_edge)
axis.add_patch(ellipse_fill)
if sampled_values is not None:
sampled_values = sampled_values.copy()[:, :2]
if scipy.sparse.issparse(sampled_values):
sampled_values = sampled_values.A
sample_colour_map = seaborn.blend_palette(("white", "purple"),
as_cmap=True)
x_limits = axis.get_xlim()
y_limits = axis.get_ylim()
axis.hexbin(sampled_values[:, 0],
sampled_values[:, 1],
gridsize=75,
cmap=sample_colour_map,
linewidths=0.,
edgecolors="none",
zorder=-100)
axis.set_xlim(x_limits)
axis.set_ylim(y_limits)
# Reset marker size
style.reset_plot_look()
return figure, figure_name
def _setup_model(data_set,
model_type=None,
latent_size=None,
hidden_sizes=None,
number_of_importance_samples=None,
number_of_monte_carlo_samples=None,
inference_architecture=None,
latent_distribution=None,
number_of_classes=None,
parameterise_latent_posterior=False,
prior_probabilities_method=None,
generative_architecture=None,
reconstruction_distribution=None,
number_of_reconstruction_classes=None,
count_sum=None,
proportion_of_free_nats_for_y_kl_divergence=None,
minibatch_normalisation=None,
batch_correction=None,
dropout_keep_probabilities=None,
number_of_warm_up_epochs=None,
kl_weight=None,
models_directory=None):
if model_type is None:
model_type = defaults["model"]["type"]
if batch_correction is None:
batch_correction = defaults["model"]["batch_correction"]
feature_size = data_set.number_of_features
number_of_batches = data_set.number_of_batches
if not data_set.has_batches:
batch_correction = False
if normalise_string(model_type) == "vae":
model = VariationalAutoencoder(
feature_size=feature_size,
latent_size=latent_size,
hidden_sizes=hidden_sizes,
number_of_monte_carlo_samples=number_of_monte_carlo_samples,
number_of_importance_samples=number_of_importance_samples,
inference_architecture=inference_architecture,
latent_distribution=latent_distribution,
number_of_latent_clusters=number_of_classes,
parameterise_latent_posterior=parameterise_latent_posterior,
generative_architecture=generative_architecture,
reconstruction_distribution=reconstruction_distribution,
number_of_reconstruction_classes=number_of_reconstruction_classes,
minibatch_normalisation=minibatch_normalisation,
batch_correction=batch_correction,
number_of_batches=number_of_batches,
dropout_keep_probabilities=dropout_keep_probabilities,
count_sum=count_sum,
number_of_warm_up_epochs=number_of_warm_up_epochs,
kl_weight=kl_weight,
log_directory=models_directory)
elif normalise_string(model_type) == "gmvae":
prior_probabilities_method_for_model = prior_probabilities_method
if prior_probabilities_method == "uniform":
prior_probabilities = None
elif prior_probabilities_method == "infer":
prior_probabilities_method_for_model = "custom"
prior_probabilities = data_set.class_probabilities
else:
prior_probabilities = None
model = GaussianMixtureVariationalAutoencoder(
feature_size=feature_size,
latent_size=latent_size,
hidden_sizes=hidden_sizes,
number_of_monte_carlo_samples=number_of_monte_carlo_samples,
number_of_importance_samples=number_of_importance_samples,
prior_probabilities_method=prior_probabilities_method_for_model,
prior_probabilities=prior_probabilities,
latent_distribution=latent_distribution,
number_of_latent_clusters=number_of_classes,
proportion_of_free_nats_for_y_kl_divergence=(
proportion_of_free_nats_for_y_kl_divergence),
reconstruction_distribution=reconstruction_distribution,
number_of_reconstruction_classes=number_of_reconstruction_classes,
minibatch_normalisation=minibatch_normalisation,
batch_correction=batch_correction,
number_of_batches=number_of_batches,
dropout_keep_probabilities=dropout_keep_probabilities,
count_sum=count_sum,
number_of_warm_up_epochs=number_of_warm_up_epochs,
kl_weight=kl_weight,
log_directory=models_directory)
else:
raise ValueError("Model type not found: `{}`.".format(model_type))
return model
def evaluate(data_set_file_or_name,
data_format=None,
data_directory=None,
map_features=None,
feature_selection=None,
example_filter=None,
noisy_preprocessing_methods=None,
preprocessing_methods=None,
split_data_set=None,
splitting_method=None,
splitting_fraction=None,
model_type=None,
latent_size=None,
hidden_sizes=None,
number_of_importance_samples=None,
number_of_monte_carlo_samples=None,
inference_architecture=None,
latent_distribution=None,
number_of_classes=None,
parameterise_latent_posterior=False,
prior_probabilities_method=None,
generative_architecture=None,
reconstruction_distribution=None,
number_of_reconstruction_classes=None,
count_sum=None,
proportion_of_free_nats_for_y_kl_divergence=None,
minibatch_normalisation=None,
batch_correction=None,
dropout_keep_probabilities=None,
number_of_warm_up_epochs=None,
kl_weight=None,
minibatch_size=None,
run_id=None,
models_directory=None,
included_analyses=None,
analysis_level=None,
decomposition_methods=None,
highlight_feature_indices=None,
export_options=None,
analyses_directory=None,
evaluation_set_kind=None,
sample_size=None,
prediction_method=None,
prediction_training_set_kind=None,
model_versions=None,
**keyword_arguments):
"""Evaluate model on data set."""
if split_data_set is None:
split_data_set = defaults["data"]["split_data_set"]
if splitting_method is None:
splitting_method = defaults["data"]["splitting_method"]
if splitting_fraction is None:
splitting_fraction = defaults["data"]["splitting_fraction"]
if models_directory is None:
models_directory = defaults["models"]["directory"]
if evaluation_set_kind is None:
evaluation_set_kind = defaults["evaluation"]["data_set_name"]
if sample_size is None:
sample_size = defaults["models"]["sample_size"]
if prediction_method is None:
prediction_method = defaults["evaluation"]["prediction_method"]
if prediction_training_set_kind is None:
prediction_training_set_kind = defaults["evaluation"][
"prediction_training_set_kind"]
if model_versions is None:
model_versions = defaults["evaluation"]["model_versions"]
if analyses_directory is None:
analyses_directory = defaults["analyses"]["directory"]
evaluation_set_kind = normalise_string(evaluation_set_kind)
prediction_training_set_kind = normalise_string(
prediction_training_set_kind)
model_versions = parse_model_versions(model_versions)
print(title("Data"))
binarise_values = False
if reconstruction_distribution == "bernoulli":
if noisy_preprocessing_methods:
if noisy_preprocessing_methods[-1] != "binarise":
noisy_preprocessing_methods.append("binarise")
else:
binarise_values = True
data_set = DataSet(data_set_file_or_name,
data_format=data_format,
directory=data_directory,
map_features=map_features,
feature_selection=feature_selection,
example_filter=example_filter,
preprocessing_methods=preprocessing_methods,
binarise_values=binarise_values,
noisy_preprocessing_methods=noisy_preprocessing_methods)
if not split_data_set or evaluation_set_kind == "full":
data_set.load()
if split_data_set:
training_set, validation_set, test_set = data_set.split(
method=splitting_method, fraction=splitting_fraction)
data_subsets = [data_set, training_set, validation_set, test_set]
for data_subset in data_subsets:
clear_data_subset = True
if data_subset.kind == evaluation_set_kind:
evaluation_set = data_subset
clear_data_subset = False
if data_subset.kind == prediction_training_set_kind:
prediction_training_set = data_subset
clear_data_subset = False
if clear_data_subset:
data_subset.clear()
else:
splitting_method = None
splitting_fraction = None
evaluation_set = data_set
prediction_training_set = data_set
evaluation_subset_indices = indices_for_evaluation_subset(evaluation_set)
models_directory = build_directory_path(
models_directory,
data_set=evaluation_set,
splitting_method=splitting_method,
splitting_fraction=splitting_fraction)
analyses_directory = build_directory_path(
analyses_directory,
data_set=evaluation_set,
splitting_method=splitting_method,
splitting_fraction=splitting_fraction)
print(title("Model"))
if number_of_classes is None:
if evaluation_set.has_labels:
number_of_classes = (evaluation_set.number_of_classes -
evaluation_set.number_of_excluded_classes)
model = _setup_model(
data_set=evaluation_set,
model_type=model_type,
latent_size=latent_size,
hidden_sizes=hidden_sizes,
number_of_importance_samples=number_of_importance_samples,
number_of_monte_carlo_samples=number_of_monte_carlo_samples,
inference_architecture=inference_architecture,
latent_distribution=latent_distribution,
number_of_classes=number_of_classes,
parameterise_latent_posterior=parameterise_latent_posterior,
prior_probabilities_method=prior_probabilities_method,
generative_architecture=generative_architecture,
reconstruction_distribution=reconstruction_distribution,
number_of_reconstruction_classes=number_of_reconstruction_classes,
count_sum=count_sum,
proportion_of_free_nats_for_y_kl_divergence=(
proportion_of_free_nats_for_y_kl_divergence),
minibatch_normalisation=minibatch_normalisation,
batch_correction=batch_correction,
dropout_keep_probabilities=dropout_keep_probabilities,
number_of_warm_up_epochs=number_of_warm_up_epochs,
kl_weight=kl_weight,
models_directory=models_directory)
if ("best_model" in model_versions
and not better_model_exists(model, run_id=run_id)):
model_versions.remove("best_model")
if ("early_stopping" in model_versions
and not model_stopped_early(model, run_id=run_id)):
model_versions.remove("early_stopping")
print(subtitle("Analysis"))
analyses.analyse_model(model=model,
run_id=run_id,
included_analyses=included_analyses,
analysis_level=analysis_level,
export_options=export_options,
analyses_directory=analyses_directory)
print(title("Results"))
print("Evaluation set: {} set.".format(evaluation_set.kind))
print("Model version{}: {}.".format(
"" if len(model_versions) == 1 else "s",
enumerate_strings([v.replace("_", " ") for v in model_versions],
conjunction="and")))
if prediction_method:
prediction_specifications = PredictionSpecifications(
method=prediction_method,
number_of_clusters=number_of_classes,
training_set_kind=prediction_training_set.kind)
print("Prediction method: {}.".format(
prediction_specifications.method))
print("Number of clusters: {}.".format(
prediction_specifications.number_of_clusters))
print("Prediction training set: {} set.".format(
prediction_specifications.training_set_kind))
print()
for model_version in model_versions:
use_best_model = False
use_early_stopping_model = False
if model_version == "best_model":
use_best_model = True
elif model_version == "early_stopping":
use_early_stopping_model = True
print(subtitle(model_version.replace("_", " ").capitalize()))
print(
heading("{} evaluation".format(
model_version.replace("_", "-").capitalize())))
(transformed_evaluation_set, reconstructed_evaluation_set,
latent_evaluation_sets) = model.evaluate(
evaluation_set=evaluation_set,
evaluation_subset_indices=evaluation_subset_indices,
minibatch_size=minibatch_size,
run_id=run_id,
use_best_model=use_best_model,
use_early_stopping_model=use_early_stopping_model,
output_versions="all")
print()
if sample_size:
print(
heading("{} sampling".format(
model_version.replace("_", "-").capitalize())))
sample_reconstruction_set, __ = model.sample(
sample_size=sample_size,
minibatch_size=minibatch_size,
run_id=run_id,
use_best_model=use_best_model,
use_early_stopping_model=use_early_stopping_model)
print()
else:
sample_reconstruction_set = None
if prediction_method:
print(
heading("{} prediction".format(
model_version.replace("_", "-").capitalize())))
latent_prediction_training_sets = model.evaluate(
evaluation_set=prediction_training_set,
minibatch_size=minibatch_size,
run_id=run_id,
use_best_model=use_best_model,
use_early_stopping_model=use_early_stopping_model,
output_versions="latent",
log_results=False)
print()
cluster_ids, predicted_labels, predicted_superset_labels = (
predict_labels(
training_set=latent_prediction_training_sets["z"],
evaluation_set=latent_evaluation_sets["z"],
specifications=prediction_specifications))
evaluation_set_versions = [
transformed_evaluation_set, reconstructed_evaluation_set
] + list(latent_evaluation_sets.values())
for evaluation_set_version in evaluation_set_versions:
evaluation_set_version.update_predictions(
prediction_specifications=prediction_specifications,
predicted_cluster_ids=cluster_ids,
predicted_labels=predicted_labels,
predicted_superset_labels=predicted_superset_labels)
print()
print(
heading("{} analysis".format(
model_version.replace("_", "-").capitalize())))
analyses.analyse_results(
evaluation_set=transformed_evaluation_set,
reconstructed_evaluation_set=reconstructed_evaluation_set,
latent_evaluation_sets=latent_evaluation_sets,
model=model,
run_id=run_id,
sample_reconstruction_set=sample_reconstruction_set,
decomposition_methods=decomposition_methods,
evaluation_subset_indices=evaluation_subset_indices,
highlight_feature_indices=highlight_feature_indices,
best_model=use_best_model,
early_stopping=use_early_stopping_model,
included_analyses=included_analyses,
analysis_level=analysis_level,
export_options=export_options,
analyses_directory=analyses_directory)
return 0
def decompose(values,
other_value_sets={},
centroids={},
method=None,
number_of_components=None,
random=False):
if method is None:
method = defaults["decomposition_method"]
method = proper_string(normalise_string(method),
DECOMPOSITION_METHOD_NAMES)
if number_of_components is None:
number_of_components = defaults["decomposition_dimensionality"]
other_values_provided_as_dictionary = True
if other_value_sets is not None and not isinstance(other_value_sets, dict):
other_value_sets["unknown"] = other_value_sets
other_values_provided_as_dictionary = False
if random:
random_state = None
else:
random_state = 42
if method == "PCA":
if (values.shape[1] <= MAXIMUM_FEATURE_SIZE_FOR_NORMAL_PCA
and not scipy.sparse.issparse(values)):
model = PCA(n_components=number_of_components)
else:
model = IncrementalPCA(n_components=number_of_components,
batch_size=100)
elif method == "SVD":
model = TruncatedSVD(n_components=number_of_components)
elif method == "ICA":
model = FastICA(n_components=number_of_components)
elif method == "t-SNE":
if number_of_components < 4:
tsne_method = "barnes_hut"
else:
tsne_method = "exact"
model = TSNE(n_components=number_of_components,
method=tsne_method,
random_state=random_state)
else:
raise ValueError("Method `{}` not found.".format(method))
values_decomposed = model.fit_transform(values)
if other_value_sets and method != "t-SNE":
other_value_sets_decomposed = {}
for other_set_name, other_values in other_value_sets.items():
if other_values is not None:
other_value_decomposed = model.transform(other_values)
else:
other_value_decomposed = None
other_value_sets_decomposed[other_set_name] = (
other_value_decomposed)
else:
other_value_sets_decomposed = None
if other_value_sets_decomposed and not other_values_provided_as_dictionary:
other_value_sets_decomposed = other_value_sets_decomposed["unknown"]
# Only supports centroids without data sets as top levels
if centroids is not None and method == "PCA":
if "means" in centroids:
centroids = {"unknown": centroids}
components = model.components_
centroids_decomposed = {}
for distribution, distribution_centroids in centroids.items():
if distribution_centroids:
centroids_distribution_decomposed = {}
for parameter, parameter_values in (
distribution_centroids.items()):
if parameter == "means":
shape = numpy.array(parameter_values.shape)
original_dimension = shape[-1]
reshaped_parameter_values = parameter_values.reshape(
-1, original_dimension)
decomposed_parameter_values = model.transform(
reshaped_parameter_values)
shape[-1] = number_of_components
new_parameter_values = (
decomposed_parameter_values.reshape(shape))
elif parameter == "covariance_matrices":
shape = numpy.array(parameter_values.shape)
original_dimension = shape[-1]
reshaped_parameter_values = parameter_values.reshape(
-1, original_dimension, original_dimension)
n_centroids = reshaped_parameter_values.shape[0]
decomposed_parameter_values = numpy.empty(
shape=(n_centroids, 2, 2))
for i in range(n_centroids):
decomposed_parameter_values[i] = (
components @ reshaped_parameter_values[i]
@ components.T)
shape[-2:] = number_of_components
new_parameter_values = (
decomposed_parameter_values.reshape(shape))
else:
new_parameter_values = parameter_values
centroids_distribution_decomposed[parameter] = (
new_parameter_values)
centroids_decomposed[distribution] = (
centroids_distribution_decomposed)
else:
centroids_decomposed[distribution] = None
if "unknown" in centroids_decomposed:
centroids_decomposed = centroids_decomposed["unknown"]
else:
centroids_decomposed = None
output = [values_decomposed]
if other_value_sets != {}:
output.append(other_value_sets_decomposed)
if centroids != {}:
output.append(centroids_decomposed)
return output
def analyse_centroid_probabilities(centroids,
name=None,
analysis_level=None,
export_options=None,
analyses_directory=None):
if name:
name = normalise_string(name)
if analysis_level is None:
analysis_level = defaults["analyses"]["analysis_level"]
if analyses_directory is None:
analyses_directory = defaults["analyses"]["directory"]
print("Plotting centroid probabilities.")
plot_time_start = time()
posterior_probabilities = None
prior_probabilities = None
if "posterior" in centroids and centroids["posterior"]:
posterior_probabilities = centroids["posterior"]["probabilities"]
n_centroids = len(posterior_probabilities)
if "prior" in centroids and centroids["prior"]:
prior_probabilities = centroids["prior"]["probabilities"]
n_centroids = len(prior_probabilities)
centroids_palette = style.darker_palette(n_centroids)
x_label = "$k$"
if prior_probabilities is not None:
if posterior_probabilities is not None:
y_label = _axis_label_for_symbol(
symbol="\\pi",
distribution=normalise_string("posterior"),
suffix="^k")
if name:
plot_name = [name, "posterior", "prior"]
else:
plot_name = ["posterior", "prior"]
else:
y_label = _axis_label_for_symbol(
symbol="\\pi",
distribution=normalise_string("prior"),
suffix="^k")
if name:
plot_name = [name, "prior"]
else:
plot_name = "prior"
elif posterior_probabilities is not None:
y_label = _axis_label_for_symbol(
symbol="\\pi",
distribution=normalise_string("posterior"),
suffix="^k")
if name:
plot_name = [name, "posterior"]
else:
plot_name = "posterior"
figure, figure_name = figures.plot_probabilities(posterior_probabilities,
prior_probabilities,
x_label=x_label,
y_label=y_label,
palette=centroids_palette,
uniform=False,
name=plot_name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=analyses_directory)
plot_duration = time() - plot_time_start
print("Centroid probabilities plotted and saved ({}).".format(
format_duration(plot_duration)))
def analyse_decompositions(data_sets,
other_data_sets=None,
centroids=None,
colouring_data_set=None,
sampled_data_set=None,
decomposition_methods=None,
highlight_feature_indices=None,
symbol=None,
title="data set",
specifier=None,
analysis_level=None,
export_options=None,
analyses_directory=None):
if analysis_level is None:
analysis_level = defaults["analyses"]["analysis_level"]
centroids_original = centroids
if isinstance(data_sets, dict):
data_sets = list(data_sets.values())
if not isinstance(data_sets, (list, tuple)):
data_sets = [data_sets]
if other_data_sets is None:
other_data_sets = [None] * len(data_sets)
elif not isinstance(other_data_sets, (list, tuple)):
other_data_sets = [other_data_sets]
if len(data_sets) != len(other_data_sets):
raise ValueError(
"Lists of data sets and alternative data sets do not have the "
"same length.")
specification = None
base_symbol = symbol
original_title = title
if decomposition_methods is None:
decomposition_methods = [defaults["decomposition_method"]]
elif not isinstance(decomposition_methods, (list, tuple)):
decomposition_methods = [decomposition_methods]
else:
decomposition_methods = decomposition_methods.copy()
decomposition_methods.insert(0, None)
if highlight_feature_indices is None:
highlight_feature_indices = defaults["analyses"][
"highlight_feature_indices"]
elif not isinstance(highlight_feature_indices, (list, tuple)):
highlight_feature_indices = [highlight_feature_indices]
else:
highlight_feature_indices = highlight_feature_indices.copy()
if analyses_directory is None:
analyses_directory = defaults["analyses"]["directory"]
for data_set, other_data_set in zip(data_sets, other_data_sets):
if data_set.values.shape[1] <= 1:
continue
title = original_title
name = normalise_string(title)
if specifier:
specification = specifier(data_set)
if specification:
name += "-" + str(specification)
title += " for " + specification
title += " set"
if not colouring_data_set:
colouring_data_set = data_set
if data_set.version in ["z", "z1"]:
centroids = copy.deepcopy(centroids_original)
else:
centroids = None
if other_data_set:
title = "{} set values in {}".format(other_data_set.version, title)
name = other_data_set.version + "-" + name
decompositions_directory = os.path.join(analyses_directory, name)
for decomposition_method in decomposition_methods:
other_values = None
sampled_values = None
if other_data_set:
other_values = other_data_set.values
if sampled_data_set:
sampled_values = sampled_data_set.values
if not decomposition_method:
if data_set.number_of_features == 2:
values_decomposed = data_set.values
other_values_decomposed = other_values
sampled_values_decomposed = sampled_values
centroids_decomposed = centroids
else:
continue
else:
decomposition_method = proper_string(
decomposition_method, DECOMPOSITION_METHOD_NAMES)
values_decomposed = data_set.values
other_values_decomposed = other_values
sampled_values_decomposed = sampled_values
centroids_decomposed = centroids
other_value_sets_decomposed = {}
if other_values is not None:
other_value_sets_decomposed["other"] = other_values
if sampled_values is not None:
other_value_sets_decomposed["sampled"] = sampled_values
if not other_value_sets_decomposed:
other_value_sets_decomposed = None
if decomposition_method == "t-SNE":
if (data_set.number_of_examples >
MAXIMUM_NUMBER_OF_EXAMPLES_FOR_TSNE):
print(
"The number of examples for {}".format(title),
"is too large to decompose it",
"using {}. Skipping.".format(decomposition_method))
print()
continue
elif (data_set.number_of_features >
MAXIMUM_NUMBER_OF_FEATURES_FOR_TSNE):
number_of_pca_components_before_tsne = min(
MAXIMUM_NUMBER_OF_PCA_COMPONENTS_BEFORE_TSNE,
data_set.number_of_examples - 1)
print(
"The number of features for {}".format(title),
"is too large to decompose it",
"using {} in due time.".format(
decomposition_method))
print("Decomposing {} to {} components using PCA "
"beforehand.".format(
title, number_of_pca_components_before_tsne))
decompose_time_start = time()
(values_decomposed, other_value_sets_decomposed,
centroids_decomposed) = decompose(
values_decomposed,
other_value_sets=other_value_sets_decomposed,
centroids=centroids_decomposed,
method="pca",
number_of_components=(
number_of_pca_components_before_tsne))
decompose_duration = time() - decompose_time_start
print("{} pre-decomposed ({}).".format(
capitalise_string(title),
format_duration(decompose_duration)))
else:
if scipy.sparse.issparse(values_decomposed):
values_decomposed = values_decomposed.A
if scipy.sparse.issparse(other_values_decomposed):
other_values_decomposed = other_values_decomposed.A
if scipy.sparse.issparse(sampled_values_decomposed):
sampled_values_decomposed = (
sampled_values_decomposed.A)
print("Decomposing {} using {}.".format(
title, decomposition_method))
decompose_time_start = time()
(values_decomposed, other_value_sets_decomposed,
centroids_decomposed) = decompose(
values_decomposed,
other_value_sets=other_value_sets_decomposed,
centroids=centroids_decomposed,
method=decomposition_method,
number_of_components=2)
decompose_duration = time() - decompose_time_start
print("{} decomposed ({}).".format(
capitalise_string(title),
format_duration(decompose_duration)))
print()
if other_value_sets_decomposed:
other_values_decomposed = other_value_sets_decomposed.get(
"other")
sampled_values_decomposed = (
other_value_sets_decomposed.get("sampled"))
if base_symbol:
symbol = base_symbol
else:
symbol = specification
x_label = _axis_label_for_symbol(
symbol=symbol,
coordinate=1,
decomposition_method=decomposition_method,
)
y_label = _axis_label_for_symbol(
symbol=symbol,
coordinate=2,
decomposition_method=decomposition_method,
)
figure_labels = {
"title": decomposition_method,
"x label": x_label,
"y label": y_label
}
if other_data_set:
plot_values_decomposed = other_values_decomposed
else:
plot_values_decomposed = values_decomposed
if plot_values_decomposed is None:
print("No values to plot.\n")
return
print("Plotting {}{}.".format(
"decomposed " if decomposition_method else "", title))
# No colour-coding
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} plotted and saved ({}).".format(
capitalise_string(title), format_duration(plot_duration)))
# Samples
if sampled_data_set:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
centroids=centroids_decomposed,
sampled_values=sampled_values_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} (with samples) plotted and saved ({}).".format(
capitalise_string(title), format_duration(plot_duration)))
# Labels
if colouring_data_set.labels is not None:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="labels",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} (with labels) plotted and saved ({}).".format(
capitalise_string(title), format_duration(plot_duration)))
# Superset labels
if colouring_data_set.superset_labels is not None:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="superset labels",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" "
"{} (with superset labels) plotted and saved ({}).".
format(capitalise_string(title),
format_duration(plot_duration)))
# For each class
if analysis_level == "extensive":
if colouring_data_set.number_of_classes <= 10:
plot_time_start = time()
for class_name in colouring_data_set.class_names:
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="class",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
class_name=class_name,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(
figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(
" {} (for each class) plotted and saved ({}).".
format(capitalise_string(title),
format_duration(plot_duration)))
if (colouring_data_set.superset_labels is not None
and data_set.number_of_superset_classes <= 10):
plot_time_start = time()
for superset_class_name in (
colouring_data_set.superset_class_names):
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="superset class",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
class_name=superset_class_name,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(
figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} (for each superset class) plotted and "
"saved ({}).".format(
capitalise_string(title),
format_duration(plot_duration)))
# Batches
if colouring_data_set.has_batches:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="batches",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name,
)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" "
"{} (with batches) plotted and saved ({}).".format(
capitalise_string(title),
format_duration(plot_duration)))
# Cluster IDs
if colouring_data_set.has_predicted_cluster_ids:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="predicted cluster IDs",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name,
)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(
" "
"{} (with predicted cluster IDs) plotted and saved ({}).".
format(capitalise_string(title),
format_duration(plot_duration)))
# Predicted labels
if colouring_data_set.has_predicted_labels:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="predicted labels",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name,
)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" "
"{} (with predicted labels) plotted and saved ({}).".
format(capitalise_string(title),
format_duration(plot_duration)))
if colouring_data_set.has_predicted_superset_labels:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="predicted superset labels",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name,
)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(
" {} (with predicted superset labels) plotted and saved"
" ({}).".format(capitalise_string(title),
format_duration(plot_duration)))
# Count sum
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="count sum",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} (with count sum) plotted and saved ({}).".format(
capitalise_string(title), format_duration(plot_duration)))
# Features
for feature_index in highlight_feature_indices:
plot_time_start = time()
figure, figure_name = figures.plot_values(
plot_values_decomposed,
colour_coding="feature",
colouring_data_set=colouring_data_set,
centroids=centroids_decomposed,
feature_index=feature_index,
figure_labels=figure_labels,
example_tag=data_set.tags["example"],
name=name)
figures.save_figure(figure=figure,
name=figure_name,
options=export_options,
directory=decompositions_directory)
plot_duration = time() - plot_time_start
print(" {} (with {}) plotted and saved ({}).".format(
capitalise_string(title),
data_set.feature_names[feature_index],
format_duration(plot_duration)))
print()
def split_data_set(data_dictionary, method=None, fraction=None):
if method is None:
method = defaults["data"]["splitting_method"]
if fraction is None:
fraction = defaults["data"]["splitting_fraction"]
print("Splitting data set.")
start_time = time()
if method == "default":
if "split indices" in data_dictionary:
method = "indices"
else:
method = "random"
method = normalise_string(method)
n = data_dictionary["values"].shape[0]
random_state = numpy.random.RandomState(42)
if method in ["random", "sequential"]:
n_training_validation = int(fraction * n)
n_training = int(fraction * n_training_validation)
if method == "random":
indices = random_state.permutation(n)
else:
indices = numpy.arange(n)
training_indices = indices[:n_training]
validation_indices = indices[n_training:n_training_validation]
test_indices = indices[n_training_validation:]
elif method == "indices":
split_indices = data_dictionary["split indices"]
training_indices = split_indices["training"]
test_indices = split_indices["test"]
if "validation" in split_indices:
validation_indices = split_indices["validation"]
else:
n_training_validation = training_indices.stop
n_all = test_indices.stop
n_training = n_training_validation - (
n_all - n_training_validation)
training_indices = slice(n_training)
validation_indices = slice(n_training, n_training_validation)
elif method == "macosko":
values = data_dictionary["values"]
minimum_number_of_non_zero_elements = 900
number_of_non_zero_elements = (values != 0).sum(axis=1)
training_indices = numpy.nonzero(
number_of_non_zero_elements > minimum_number_of_non_zero_elements
)[0]
test_validation_indices = numpy.nonzero(
number_of_non_zero_elements <= minimum_number_of_non_zero_elements
)[0]
random_state.shuffle(test_validation_indices)
n_validation_test = len(test_validation_indices)
n_validation = int((1 - fraction) * n_validation_test)
validation_indices = test_validation_indices[:n_validation]
test_indices = test_validation_indices[n_validation:]
else:
raise ValueError("Splitting method `{}` not found.".format(method))
split_data_dictionary = {
"training set": {
"values": data_dictionary["values"][training_indices],
"preprocessed values": None,
"binarised values": None,
"labels": None,
"example names":
data_dictionary["example names"][training_indices],
"batch indices": None
},
"validation set": {
"values": data_dictionary["values"][validation_indices],
"preprocessed values": None,
"binarised values": None,
"labels": None,
"example names":
data_dictionary["example names"][validation_indices],
"batch indices": None
},
"test set": {
"values": data_dictionary["values"][test_indices],
"preprocessed values": None,
"binarised values": None,
"labels": None,
"example names": data_dictionary["example names"][test_indices],
"batch indices": None
},
"feature names": data_dictionary["feature names"],
"class names": data_dictionary["class names"]
}
if "labels" in data_dictionary and data_dictionary["labels"] is not None:
split_data_dictionary["training set"]["labels"] = (
data_dictionary["labels"][training_indices])
split_data_dictionary["validation set"]["labels"] = (
data_dictionary["labels"][validation_indices])
split_data_dictionary["test set"]["labels"] = (
data_dictionary["labels"][test_indices])
if ("preprocessed values" in data_dictionary
and data_dictionary["preprocessed values"] is not None):
split_data_dictionary["training set"]["preprocessed values"] = (
data_dictionary["preprocessed values"][training_indices])
split_data_dictionary["validation set"]["preprocessed values"] = (
data_dictionary["preprocessed values"][validation_indices])
split_data_dictionary["test set"]["preprocessed values"] = (
data_dictionary["preprocessed values"][test_indices])
if ("binarised values" in data_dictionary
and data_dictionary["binarised values"] is not None):
split_data_dictionary["training set"]["binarised values"] = (
data_dictionary["binarised values"][training_indices])
split_data_dictionary["validation set"]["binarised values"] = (
data_dictionary["binarised values"][validation_indices])
split_data_dictionary["test set"]["binarised values"] = (
data_dictionary["binarised values"][test_indices])
if ("batch indices" in data_dictionary
and data_dictionary["batch indices"] is not None):
split_data_dictionary["training set"]["batch indices"] = (
data_dictionary["batch indices"][training_indices])
split_data_dictionary["validation set"]["batch indices"] = (
data_dictionary["batch indices"][validation_indices])
split_data_dictionary["test set"]["batch indices"] = (
data_dictionary["batch indices"][test_indices])
duration = time() - start_time
print("Data set split ({}).".format(format_duration(duration)))
return split_data_dictionary