Skip to content

Commit

Permalink
Merge pull request #67 from ealcobaca/unsupervised
Browse files Browse the repository at this point in the history 
Support for metafeature extraction in unsupervised tasks
  • Loading branch information
@ealcobaca
ealcobaca committed Mar 6, 2020
2 parents dc64e5f + aeea3e8 commit f53221a
Show file tree
Hide file tree
Showing 11 changed files with 356 additions and 172 deletions.
173 changes: 142 additions & 31 deletions pymfe/_internal.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,7 @@
import inspect
import collections
import warnings
import shutil
import time
import sys
import re
Expand Down Expand Up @@ -171,8 +172,9 @@
TypeMtdTuple = t.Tuple[str, t.Callable[[], t.Any]]
"""Type annotation which describes the a metafeature method tuple."""

TypeExtMtdTuple = t.Tuple[str, t.Callable[[], t.Any], t.Sequence]
"""Type annotation which extends TypeMtdTuple with extra field (``Args``)"""
TypeExtMtdTuple = t.Tuple[str, t.Callable[[], t.Any],
t.Tuple[str, ...], t.Tuple[str, ...]]
"""Type annotation which extends TypeMtdTuple with extra fields."""

_TYPE_NUMERIC = (
int,
Expand All @@ -190,6 +192,13 @@
"""Typing alias of generic numeric types for static code checking."""


VERBOSE_BLOCK_MID_SYMBOL = "|"

VERBOSE_BLOCK_END_SYMBOL = "."

VERBOSE_WARNING_SYMBOL = "*"


def warning_format(message: str,
category: t.Type[Warning],
filename: str,
Expand All @@ -212,7 +221,7 @@ def warning_format(message: str,
str: formated warning message.
"""
# pylint: disable=W0613
return "Warning: {}\n".format(message)
return " {} Warning: {}\n".format(VERBOSE_WARNING_SYMBOL, message)


warnings.formatwarning = warning_format
Expand Down Expand Up @@ -450,22 +459,38 @@ def _preprocess_iterable_arg(
return list(map(str.lower, set(values)))


def _extract_mtd_args(ft_mtd_callable: t.Callable) -> t.Tuple[str, ...]:
def _extract_mtd_args(ft_mtd_callable: t.Callable,
) -> t.Tuple[t.Tuple[str, ...], t.Tuple[str, ...]]:
"""Extracts arguments from given method.
Args:
ft_mtd_callable (:obj:`callable`): a callable related to a feature
extraction method.
Returns:
list: containing the name of arguments of ``ft_mtd_callable``.
tuple of tuples:
The first tuple contains the name of arguments of the given
method. The second tuple contains the name of the mandatory
arguments (i.e., arguments that does not have a default value
defined.)
Raises:
TypeError: if ``ft_mtd_callable`` is not a valid callable.
"""
ft_mtd_signature = inspect.signature(ft_mtd_callable)
mtd_callable_args = tuple(ft_mtd_signature.parameters.keys())
return mtd_callable_args
ft_mtd_signature = inspect.signature(ft_mtd_callable).parameters
mtd_callable_args = tuple(ft_mtd_signature.keys())

# pylint: disable=W0212
# 'Access to a protected member _empty of a client class'
_empty = inspect._empty # type: ignore

mandatory_args = tuple(
cur_param
for cur_param in ft_mtd_signature.keys()
if ft_mtd_signature[cur_param].default is _empty
)

return mtd_callable_args, mandatory_args


def summarize(
Expand Down Expand Up @@ -569,6 +594,7 @@ def get_feat_value(

def build_mtd_kwargs(mtd_name: str,
mtd_args: t.Iterable[str],
mtd_mandatory: t.Iterable[str],
inner_custom_args: t.Optional[t.Dict[str, t.Any]] = None,
user_custom_args: t.Optional[t.Dict[str, t.Any]] = None,
precomp_args: t.Optional[t.Dict[str, t.Any]] = None,
Expand All @@ -581,6 +607,10 @@ def build_mtd_kwargs(mtd_name: str,
mtd_args (:obj:`iterable` of :obj:`str`): iterable containing the name
of all arguments of the callable.
mtd_mandatory (:obj:`iterable` of :obj:`bool`): iterable containing
the name of all method's mandatory arguments (i.e., arguments
that does not have a default value defined.)
inner_custom_args (:obj:`dict`, optional): custom arguments for inner
usage, for example, to pass ``X``, ``y`` or other user-independent
arguments necessary for the callable. The expected format of this
Expand Down Expand Up @@ -615,16 +645,19 @@ def build_mtd_kwargs(mtd_name: str,
precomp_args = {}

combined_args = {
**user_custom_args,
**inner_custom_args,
**precomp_args,
**user_custom_args,
}

callable_args = {
custom_arg: combined_args[custom_arg]
for custom_arg in combined_args if custom_arg in mtd_args
}

if not set(mtd_mandatory).issubset(callable_args):
raise RuntimeError("Method mandatory arguments not satisfied.")

if not suppress_warnings:
unknown_arg_set = (unknown_arg
for unknown_arg in user_custom_args.keys()
Expand Down Expand Up @@ -925,12 +958,14 @@ def process_summary(
summary_func),
RuntimeWarning)
else:
summary_mtd_args = _extract_mtd_args(summary_mtd_callable)
summary_mtd_args, mandatory = _extract_mtd_args(
summary_mtd_callable)

summary_mtd_pack = (
summary_func,
summary_mtd_callable,
summary_mtd_args,
mandatory,
)

summary_methods.append(summary_mtd_pack)
Expand Down Expand Up @@ -1029,10 +1064,11 @@ def process_features(
ft_mtd_name, ft_mtd_callable = ft_mtd_tuple

if ft_mtd_name in processed_ft:
mtd_callable_args = _extract_mtd_args(ft_mtd_callable)
mtd_callable_args, mandatory = _extract_mtd_args(ft_mtd_callable)

extended_item = (*ft_mtd_tuple,
mtd_callable_args) # type: TypeExtMtdTuple
mtd_callable_args,
mandatory) # type: TypeExtMtdTuple

ft_mtd_processed.append(extended_item)
available_feat_names.append(ft_mtd_name)
Expand Down Expand Up @@ -1070,6 +1106,7 @@ def process_precomp_groups(
groups: t.Optional[t.Tuple[str, ...]] = None,
wildcard: str = "all",
suppress_warnings: bool = False,
verbose: int = 0,
custom_class_: t.Any = None,
**kwargs) -> t.Dict[str, t.Any]:
"""Process ``precomp_groups`` argument while fitting into a MFE model.
Expand Down Expand Up @@ -1098,13 +1135,21 @@ def process_precomp_groups(
not None, the given class will be used as reference to extract
the preprocomputing methods.
verbose (:obj:`int`, optional): defines the level of verbosity in
this function. If `1`, print a progress bar related to the
precomputation process. If `2` or higher, then log every step of
the precomputation process.
**kwargs: used to pass extra custom arguments to precomputation metho-
ds.
Returns:
dict: precomputed values given by ``kwargs`` using convenient methods
based in valid selected metafeature groups.
"""
# pylint: disable=R0912
# 'Too many branches (15/12) (too-many-branches)'

if groups is None:
groups = tuple()

Expand All @@ -1125,9 +1170,9 @@ def process_precomp_groups(

for unknown_precomp in unknown_groups:
warnings.warn(
"Unknown precomp_groups '{0}'. You can check available "
"metafeature groups using 'valid_groups()' method."
.format(unknown_precomp), UserWarning)
" {} Unknown precomp_groups '{}'. You can check available "
"metafeature groups using 'valid_groups()' method.".format(
VERBOSE_WARNING_SYMBOL, unknown_precomp), UserWarning)

processed_precomp_groups = tuple(
set(processed_precomp_groups).intersection(groups))
Expand All @@ -1145,39 +1190,75 @@ def process_precomp_groups(

precomp_items = {} # type: t.Dict[str, t.Any]

for precomp_mtd_tuple in precomp_mtds_filtered:
error_count = 0
_prev_precomp_len = 0

for ind, precomp_mtd_tuple in enumerate(precomp_mtds_filtered, 1):
precomp_mtd_name, precomp_mtd_callable = precomp_mtd_tuple

if verbose >= 2:
print("\nStarted precomputing '{}'.".format(precomp_mtd_name))

try:
new_precomp_vals = precomp_mtd_callable(**kwargs) # type: ignore

except (AttributeError, TypeError, ValueError) as type_err:
new_precomp_vals = {}

if not suppress_warnings:
warnings.warn("Something went wrong while "
"precomputing '{0}'. Will ignore "
warnings.warn(" {} Something went wrong while "
"precomputing '{}'. Will ignore "
"this method. Error message:\n"
"{1}.".format(precomp_mtd_name, repr(type_err)))
"{}.".format(VERBOSE_WARNING_SYMBOL,
precomp_mtd_name,
repr(type_err)))

error_count += 1

if new_precomp_vals:
precomp_items = {
**precomp_items,
**new_precomp_vals,
}

new_item_count = len(precomp_items) - _prev_precomp_len
_prev_precomp_len = len(precomp_items)

if verbose >= 2 and new_item_count > 0:
print(" {} Got {} new precomputed values.".format(
VERBOSE_BLOCK_END_SYMBOL, new_item_count))

# Update kwargs to avoid recalculations iteratively
kwargs = {
**kwargs,
**new_precomp_vals,
}

if verbose > 0:
print_verbose_progress(
cur_progress=100 * ind / len(precomp_mtds_filtered),
cur_mtf_name=precomp_mtd_name,
item_type="precomputation",
verbose=verbose)

if verbose == 1:
_t_num_cols, _ = shutil.get_terminal_size()
print("\r{:<{fill}}".format(
"Process of precomputation finished.", fill=_t_num_cols))

if verbose >= 2 and error_count > 0:
print("\nNote: can't precompute a total of {} metafeatures, "
"out of {} ({:.2f}%).".format(
error_count,
len(precomp_mtds_filtered),
100 * error_count / len(precomp_mtds_filtered)))

return precomp_items


def check_data(X: t.Union[np.ndarray, list],
y: t.Union[np.ndarray, list]
) -> t.Tuple[np.ndarray, np.ndarray]:
) -> t.Tuple[np.ndarray, t.Optional[np.ndarray]]:
"""Checks ``X`` and ``y`` data type and shape and transform it if necessary.
Args:
Expand All @@ -1197,7 +1278,7 @@ def check_data(X: t.Union[np.ndarray, list],
if not isinstance(X, (np.ndarray, list)):
raise TypeError('"X" is neither "list" nor "np.array".')

if not isinstance(y, (np.ndarray, list)):
if y is not None and not isinstance(y, (np.ndarray, list)):
raise TypeError('"y" is neither "list" nor "np.array".')

# We force numpy array to assume ``dtype=np.object`` because sometimes
Expand All @@ -1216,22 +1297,26 @@ def check_data(X: t.Union[np.ndarray, list],
if not isinstance(X, np.ndarray):
X = np.array(X, dtype=np.object)

if not isinstance(y, np.ndarray):
y = np.array(y, dtype=np.object)
if y is not None:
if not isinstance(y, np.ndarray):
y = np.array(y, dtype=np.object)

y = y.flatten()
y = y.flatten()

if len(X.shape) == 1 and X.shape[0]:
X = X.reshape(*X.shape, -1)
if X.ndim == 1:
X = X.reshape(-1, 1)

if X.shape[0] == 0 or y.shape[0] == 0:
if X.size == 0 or (y is not None and y.size == 0):
raise ValueError('Neither "X" nor "y" can be empty.')

if X.shape[0] != y.shape[0]:
raise ValueError('"X" number of rows and "y" '
"length shapes do not match.")
if y is not None:
if X.shape[0] != y.shape[0]:
raise ValueError('"X" number of rows and "y" '
"length shapes do not match.")

return np.copy(X), np.copy(y)

return np.copy(X), np.copy(y)
return np.copy(X), None


def isnumeric(
Expand Down Expand Up @@ -1660,3 +1745,29 @@ def post_processing(

if remove_groups:
kwargs.pop("groups")


def print_verbose_progress(
cur_progress: float,
cur_mtf_name: str,
item_type: str,
verbose: int = 0) -> None:
"""Print messages about extraction progress based on ``verbose``."""
if verbose >= 2:
print("Done with '{}' {} (progress of {:.2f}%)."
.format(cur_mtf_name, item_type, cur_progress))
return

_t_num_cols, _ = shutil.get_terminal_size()
_t_num_cols -= 9

if _t_num_cols <= 0:
return

_total_prog_symb = int(cur_progress * _t_num_cols / 100)

print("".join([
"\r[",
_total_prog_symb * "#",
(_t_num_cols - _total_prog_symb) * ".",
"]{:.2f}%".format(cur_progress)]), end="")

0 comments on commit f53221a

Please sign in to comment.