def normalize_scores(self, verbose=True):
"""Normalizes scores. The normalizing constant for each image is determined by
Z = mode(pixel values) / median(all modes in h5_fpath)
"""
def get_mode_in_im(im):
w = 200
hw = w / 2
rmid, cmid = int(im.shape[0] / 2), int(im.shape[1] / 2)
vmin, vmax = im.min(), im.max()
# remove saturation
pct95 = vmin + 0.95 * (vmax - vmin)
vals = [
v for v in im[rmid - hw:rmid + hw,
cmid - hw:cmid + hw].flatten() if v < pct95
]
return misc.get_mode(vals)
self.scores = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in self.h5_fpaths
}
self.normalizing_constants = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in self.h5_fpaths
}
for h5_fpath in self.h5_fpaths:
if verbose:
print(F"Basename of h5_fpath: {os.path.basename(h5_fpath)}")
for channel in self.scores[h5_fpath].keys():
mode_given_pos_tup = {}
for pos_tup in self.raw_scores[h5_fpath][channel].keys():
pos_key = hdf5tools.get_image_key(*pos_tup)
with h5py.File(h5_fpath) as f:
im = np.array(f[channel][pos_key])
mode_given_pos_tup[pos_tup] = get_mode_in_im(im)
median_of_modes = np.median(mode_given_pos_tup.values())
for pos_tup in mode_given_pos_tup.keys():
Z = mode_given_pos_tup[pos_tup] / float(median_of_modes)
self.normalizing_constants[h5_fpath][channel][pos_tup] = Z
im_scores = self.raw_scores[h5_fpath][channel][pos_tup]
self.scores[h5_fpath][channel][pos_tup] = {
read_name: im_scores[read_name] / Z
for read_name in self.get_read_names_in_image(
h5_fpath, channel, pos_tup)
}
if verbose: print
def normalize_scores_by_ref_read_names(self,
ref_read_names_given_channel,
verbose=True):
"""Normalizes scores. The normalizing constant for each image is determined by
Z = median(reference read scores) / 100
"""
self.scores = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in self.h5_fpaths
}
self.normalizing_constants = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in self.h5_fpaths
}
for h5_fpath in self.h5_fpaths:
log.debug(os.path.basename(h5_fpath))
for channel in self.scores[h5_fpath].keys():
ref_read_names = ref_read_names_given_channel[channel]
for pos_tup in self.raw_scores[h5_fpath][channel].keys():
ref_read_names_in_image = (self.get_read_names_in_image(
h5_fpath, channel, pos_tup)
& ref_read_names)
if len(ref_read_names_in_image) < 10:
print(
F"Warning: 10 > {len(ref_read_names_in_image)} reference reads in im_idx {h5_fpath}, {channel}, {pos_tup}"
)
med = np.median([
self.raw_scores[h5_fpath][channel][pos_tup][read_name]
for read_name in ref_read_names_in_image
])
Z = med / 100.0
self.normalizing_constants[h5_fpath][channel][pos_tup] = Z
im_scores = self.raw_scores[h5_fpath][channel][pos_tup]
self.scores[h5_fpath][channel][pos_tup] = {
read_name: im_scores[read_name] / Z
for read_name in self.get_read_names_in_image(
h5_fpath, channel, pos_tup)
}
def __init__(self, h5_fpaths):
"""Initialize h5_fpaths and scores. scores is a dict accessed as:
scores[h5_fpath][channel][pos_tup][read_name]
"""
self.h5_fpaths = h5_fpaths
self.raw_scores = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in h5_fpaths
}
self.scores = self.raw_scores
def build_score_given_read_name_given_channel(self):
self.score_given_read_name_in_channel = {
h5_fpath: {
channel: {}
for channel in hdf5tools.load_channel_names(h5_fpath)
}
for h5_fpath in self.h5_fpaths
}
for h5_fpath in self.h5_fpaths:
print(F"h5_fpath: {h5_fpath}")
i = 0
for channel in self.scores[h5_fpath].keys():
score_given_read_name = self.score_given_read_name_in_channel[
h5_fpath][channel]
for pos_tup in self.scores[h5_fpath][channel].keys():
for read_name, score in self.scores[h5_fpath][channel][
pos_tup].items():
score_given_read_name[read_name] = score
i += 1
