def test_sample_compositions_by_species(self):
dist_df = Utils.sample_compositions_by_species(self.spectros_dir)
#truth = pd.DataFrame.from_dict({'AMADEC' : 1, 'FORANA' : 5}, orient='index', columns=['num_samples'])
self.assertListEqual(list(dist_df.columns), ['num_samples'])
self.assertEqual(int(dist_df.loc['AMADEC']), 1)
self.assertEqual(int(dist_df.loc['FORANA']), 5)
def test_compute_num_augs_per_species(self):
aug_volumes = AugmentationGoals.MAX
sample_distrib_df = Utils.sample_compositions_by_species(
self.spectros_dir)
augs_to_do = Utils.compute_num_augs_per_species(
aug_volumes, sample_distrib_df)
self.assertEqual(augs_to_do['AMADEC'], 4)
self.assertEqual(augs_to_do['FORANA'], 0)
def __init__(self,
input_dir_path,
plot=False,
overwrite_policy=False,
aug_goals=AugmentationGoals.MEDIAN,
random_augs = False,
multiple_augs = False,):
'''
:param input_dir_path: directory holding .wav files
:type input_dir_path: str
:param plot: whether or not to plot informative chars
along the way
:type plot: bool
:param overwrite_policy: if true, don't ask each time
previously created work will be replaced
:type overwrite_policy: bool
:param aug_goals: either an AugmentationGoals member,
or a dict with a separate AugmentationGoals
for each species: {species : AugmentationGoals}
(See definition of AugmentationGoals; TENTH/MAX/MEDIAN)
:type aug_goals: {AugmentationGoals | {str : AugmentationGoals}}
:param random_augs: if this is true, will randomly choose augmentation
to use for each new sample
:type random_augs: bool
:param multiple_augs: if we want to allow multiple augmentations per sample
(e.g. time shift and volume)):
:type multiple_augs: bool
'''
self.log = LoggingService()
if not isinstance(overwrite_policy, WhenAlreadyDone):
raise TypeError(f"Overwrite policy must be a member of WhenAlreadyDone, not {type(overwrite_policy)}")
if not os.path.isabs(input_dir_path):
raise ValueError(f"Input path must be a full, absolute path; not {input_dir_path}")
self.input_dir_path = input_dir_path
self.multiple_augs = multiple_augs
self.plot = plot
self.overwrite_policy = overwrite_policy
self.species_names = Utils.find_species_names(self.input_dir_path)
# If aug_goals is not a dict mapping
# each species to an aug_goals, but just
# a single AugmentationGoals, create
# a dict from all bird species, mapping
# each to that same value:
if type(aug_goals) != dict:
aug_goals = {species : aug_goals
for species in self.species_names
}
# Get dataframe with row lables being the
# species, and one col with number of samples
# in the respective species:
# num_samples
# sp1 10
# sp2 15
# ..
self.sample_distrib_df = Utils.sample_compositions_by_species(input_dir_path,
augmented=False)
if plot:
# Plot a distribution:
self.sample_distrib_df.plot.bar()
# Build a dict with number of augmentations to do
# for each species:
self.augs_to_do = Utils.compute_num_augs_per_species(aug_goals,
self.sample_distrib_df)
# Get input dir path without trailing slash:
#**** canonical_in_path = str(Path(input_dir_path))
# Create the descriptive name of an output directory
# for the augmented samples:
if random_augs:
os.path.join(Path(input_dir_path).parent, 'augmented_samples_random')
self.output_dir_path = os.path.join(Path(input_dir_path).parent,
'augmented_samples_random')
else:
assert(self.ADD_NOISE + self.TIME_SHIFT + self.VOLUME == 1)
dir_nm = f"Augmented_samples_-{self.ADD_NOISE:.2f}n-{self.TIME_SHIFT:.2f}ts-{self.VOLUME:.2f}w"
self.output_dir_path = os.path.join(Path(input_dir_path).parent, dir_nm)
if self.multiple_augs:
self.output_dir_path += "/"
else:
# Indicate that augmentations are mutually exclusive
self.output_dir_path += "-exc/"
self.log.info(f"Results will be in {self.output_dir_path}")
Utils.create_folder(self.output_dir_path, self.overwrite_policy)
# Hide the UserWarning: PySoundFile failed. Trying audioread instead.
warnings.filterwarnings(action="ignore",
message="PySoundFile failed. Trying audioread instead.",
category=UserWarning,
module='',
lineno=0)
def __init__(self,
input_dir_path,
output_dir_path,
plot=False,
overwrite_policy=False,
aug_goals=AugmentationGoals.MEDIAN):
'''
:param input_dir_path: directory holding .png files
:type input_dir_path: str
:param output_dir_path: root of destination dir under
which each species' subdirectories will be placed.
Augmentations will be placed in those subdirs.
:type output_dir_path: str
:param plot: whether or not to plot informative charts
along the way
:type plot: bool
:param overwrite_policy: if true, don't ask each time
previously created work will be replaced
:type overwrite_policy: bool
:param aug_goals: either an AugmentationGoals member,
or a dict with a separate AugmentationGoals
for each species: {species : AugmentationGoals}
(See definition of AugmentationGoals; TENTH/MAX/MEDIAN)
:type aug_goals: {AugmentationGoals | {str : AugmentationGoals}}
'''
self.log = LoggingService()
if not isinstance(overwrite_policy, WhenAlreadyDone):
raise TypeError(
f"Overwrite policy must be a member of WhenAlreadyDone, not {type(overwrite_policy)}"
)
if not os.path.isabs(input_dir_path):
raise ValueError(
f"Input path must be a full, absolute path; not {input_dir_path}"
)
self.input_dir_path = input_dir_path
self.output_dir_path = output_dir_path
self.plot = plot
self.overwrite_policy = overwrite_policy
self.species_names = Utils.find_species_names(self.input_dir_path)
# Get dataframe with row lables being the
# species, and one col with number of samples
# in the respective species:
# num_species
# sp1 10
# sp2 15
# ..
self.sample_distrib_df = Utils.sample_compositions_by_species(
input_dir_path, augmented=False)
if plot:
# Plot a distribution:
self.sample_distrib_df.plot.bar()
# Build a dict with number of augmentations to do
# for each species:
self.augs_to_do = Utils.compute_num_augs_per_species(
aug_goals, self.sample_distrib_df)
self.log.info(f"Results will be in {self.output_dir_path}")
Utils.create_folder(self.output_dir_path, self.overwrite_policy)