def fig10(template_dir, build_dir, bound_radius=0.6, selectivity_norm=0.07138,
size_factor=1.0, show_raster=False):
name = 'fig10'
if show_raster: return Display(snapshot=name, template_dir=template_dir)
times=[2000*i for i in range(6)]
template_path, output_dir = setup(name, template_dir, build_dir)
pattern = 'jn13_figures/output/Fig10_12/*/*/*.npz'
files = lancet.FilePattern('npz_file', pattern)
file_info = lancet.FileInfo(files, 'npz_file', lancet.NumpyFile())
filter_times = file_info.dframe['time'].map(lambda x: x in times)
selection = file_info.load(file_info.dframe[filter_times])
savefig_opts = dict(transparent=True, format='svg', bbox_inches='tight', pad_inches=0)
development_info = []
for (index, row) in selection.iterrows():
time =row['time']
fname = os.path.join(output_dir, '%%s_%.1f.png' % int(time))
bounds = imagen.boundingregion.BoundingBox(radius=0.75)
sheetcoords = imagen.SheetCoordinateSystem(bounds, xdensity=98)
roi = slice(*sheetcoords.sheet2matrixidx(bound_radius, bound_radius))
# Not applying ROI as new bounds will be used
pref = get_pref(row, roi=roi)
sel = get_sel(row, roi=roi)
(cfs, coords) = row['afferent_CFs']
roi = slice(*sheetcoords.sheet2matrixidx(bound_radius, bound_radius))
# Create the combined preference/selectivity map
or_map = rasterplots.OR_map(pref,sel)
or_black = rasterplots.black_selectivity(or_map)
(dim1,dim2) = pref.shape
or_black_resized = rasterplots.resize(or_black, (dim1*4,dim2*4))
# Save the orientation map
or_black_resized.save(fname % 'OR')
# Save the row of CF images
cf_block = rasterplots.cf_image(cfs, coords, border=5, width=8, height=1,
pos=(1,4), bg=(255,255,255))
cf_block.save(fname % 'CF')
# Save the FFT of the last map
if time == times[-1]:
normalized_spectrum = analysis.power_spectrum(pref[roi,roi])
fft_im = rasterplots.greyscale(normalized_spectrum)
fft_im.save(fname % 'FFT')
# Note: ROI not applied and the values of 3 are due to historical reasons
development_info.append((time,pref[3:-3, 3:-3], np.mean(sel[3:-3,3:-3])))
ts, prefs, mean_sels = zip(*sorted(development_info))
# We only need the stabilities up till 10000
fname = os.path.join(output_dir, 'model_development.svg')
development_fig = vectorplots.map_development_plot(analysis.stability_index(prefs),
mean_sels, selectivity_norm)
development_fig.savefig(fname, **savefig_opts)
vectorplots.close([development_fig])
svg_path = os.path.join(output_dir, '%s.svg' % name)
return Display(compose.apply_template(template_path, svg_path,
size_factor=size_factor))
def fig10(template_dir,
build_dir,
bound_radius=0.6,
selectivity_norm=0.07138,
size_factor=1.0,
show_raster=False):
name = 'fig10'
if show_raster: return Display(snapshot=name, template_dir=template_dir)
times = [2000 * i for i in range(6)]
template_path, output_dir = setup(name, template_dir, build_dir)
pattern = 'jn13_figures/output/Fig10_12/*/*/*.npz'
files = lancet.FilePattern('npz_file', pattern)
file_info = lancet.FileInfo(files, 'npz_file', lancet.NumpyFile())
filter_times = file_info.dframe['time'].map(lambda x: x in times)
selection = file_info.load(file_info.dframe[filter_times])
savefig_opts = dict(transparent=True,
format='svg',
bbox_inches='tight',
pad_inches=0)
development_info = []
for (index, row) in selection.iterrows():
time = row['time']
fname = os.path.join(output_dir, '%%s_%.1f.png' % int(time))
bounds = imagen.boundingregion.BoundingBox(radius=0.75)
sheetcoords = imagen.SheetCoordinateSystem(bounds, xdensity=98)
roi = slice(*sheetcoords.sheet2matrixidx(bound_radius, bound_radius))
# Not applying ROI as new bounds will be used
pref = get_pref(row, roi=roi)
sel = get_sel(row, roi=roi)
(cfs, coords) = row['afferent_CFs']
roi = slice(*sheetcoords.sheet2matrixidx(bound_radius, bound_radius))
# Create the combined preference/selectivity map
or_map = rasterplots.OR_map(pref, sel)
or_black = rasterplots.black_selectivity(or_map)
(dim1, dim2) = pref.shape
or_black_resized = rasterplots.resize(or_black, (dim1 * 4, dim2 * 4))
# Save the orientation map
or_black_resized.save(fname % 'OR')
# Save the row of CF images
cf_block = rasterplots.cf_image(cfs,
coords,
border=5,
width=8,
height=1,
pos=(1, 4),
bg=(255, 255, 255))
cf_block.save(fname % 'CF')
# Save the FFT of the last map
if time == times[-1]:
normalized_spectrum = analysis.power_spectrum(pref[roi, roi])
fft_im = rasterplots.greyscale(normalized_spectrum)
fft_im.save(fname % 'FFT')
# Note: ROI not applied and the values of 3 are due to historical reasons
development_info.append((time, pref[3:-3,
3:-3], np.mean(sel[3:-3, 3:-3])))
ts, prefs, mean_sels = zip(*sorted(development_info))
# We only need the stabilities up till 10000
fname = os.path.join(output_dir, 'model_development.svg')
development_fig = vectorplots.map_development_plot(
analysis.stability_index(prefs), mean_sels, selectivity_norm)
development_fig.savefig(fname, **savefig_opts)
vectorplots.close([development_fig])
svg_path = os.path.join(output_dir, '%s.svg' % name)
return Display(
compose.apply_template(template_path,
svg_path,
size_factor=size_factor))
def fig06_09(template_dir, build_dir, fig, contrasts=(10,25,100),
selectivity_norm = 1.0, input_seed=102, size_factor=1.0,
kernel=analysis.gamma_metric, show_raster=False):
name = 'fig0%d' % fig
if show_raster: return Display(snapshot=name, template_dir=template_dir)
template_path, output_dir = setup('fig06_09', template_dir,
build_dir, subdir=name)
# Get the files containing the relevant data
output_name = ('Fig0%d' % fig) if fig != 6 else 'Fig05_06'
pattern = 'jn13_figures/output/%s/*%s_*/*/*.npz' % (output_name, output_name)
files = lancet.FilePattern('npz_file', pattern)
info = lancet.FileInfo(files, 'npz_file', lancet.NumpyFile())
dframe = info.dframe
# Some basic constants
max_time = max(dframe['time']); input_seeds = set(dframe['input_seed'])
all_contrasts = set(dframe['contrast']); all_times = set(dframe['time'])
savefig_opts = dict(transparent=True, format='svg', bbox_inches='tight', pad_inches=0)
# Saving all the png/svg components for the three chosen contrasts (subplot D-F)
boolarr = np.array(dframe['time'] == max_time) & np.array(dframe['input_seed'] == input_seed)
for contrast in contrasts:
fname = os.path.join(output_dir, '%%s_%d_%d.%%s' % (contrast, max_time))
contrast_slice = np.array(dframe['contrast']==contrast)
row = info.load(dframe[boolarr & contrast_slice]).iloc[0]
OR_analysis(row, template_dir, output_dir,
name='OR_c%s' % contrast, selectivity=True)
# Collecting preferences/selectivities across time/seeds for map_development plots
selectivities, stabilities = [],[]
for seed in input_seeds:
seed_slice = np.array(dframe['input_seed'] == seed)
df = info.load(dframe[contrast_slice & seed_slice])
# Ensuring the data is sorted by simulation time
sorted_rows = [df[df['time'] == time].iloc[0] for time in sorted(df['time'])]
prefs = [get_pref(row) for row in sorted_rows]
# Each selectivity is a whole map, these selectivities are averages across the map
mean_sels = [get_sel(row).mean() for row in sorted_rows]
selectivities.append(mean_sels)
stabilities.append(analysis.stability_index(prefs))
# Save the stability/selectivity plots
map_dev = vectorplots.map_development_streams(np.array(stabilities),
np.array(selectivities),
selectivity_norm = selectivity_norm)
fname = os.path.join(output_dir, 'SI_%s.svg' % contrast)
map_dev.savefig(fname,**savefig_opts)
vectorplots.close([map_dev])
# Collect the selectivity, stability and map metric across contrasts and seeds
selectivities, stabilities, metrics = [],[],[]
time_slice = np.array(dframe['time'] == max_time)
for seed in input_seeds:
seed_slice = np.array(dframe['input_seed'] == seed)
df = info.load(dframe[time_slice & seed_slice])
# Ensuring contrasts are sorted
sorted_rows = [df[df['contrast'] == contrast].iloc[0] for contrast in sorted(df['contrast'])]
missing_contrasts = set(df['contrast']) ^set(all_contrasts)
if missing_contrasts != set():
missing = ', '.join(str(el) for el in missing_contrasts)
print "Warning: Contrast values %s are missing for seed %d" % (missing, seed)
# Each selectivity is a whole map, these selectivities are averages across the map
mean_sels = [get_sel(row).mean() for row in sorted_rows]
# Computing map quality based on the pinwheel density
pinwheel_counts = [row['pinwheels'].shape[0] for row in sorted_rows]
# Pinwheel density is pinwheel count / (kmax**2)
pinwheel_densities = [(pwcount/row['kmax']**2) for (pwcount,row)
in zip(pinwheel_counts, sorted_rows)]
metrics.append([kernel(pwd) for pwd in pinwheel_densities])
selectivities.append(mean_sels)
# Obtaining the mean stability per contrast, sorted by contrast then by time
df = info.load(dframe[seed_slice])
contrast_stabilities = []
for contrast in sorted(all_contrasts):
rows = [df[(df['time']==time) & (df['contrast']==contrast)].iloc[0] for time in sorted(all_times)]
prefs = [get_pref(row) for row in rows]
mean_stability = np.mean(analysis.stability_index(prefs))
contrast_stabilities.append(mean_stability)
stabilities.append(contrast_stabilities)
# Save the contrast plot
fname = os.path.join(output_dir, 'contrast.svg')
args = (sorted(all_contrasts), np.array(selectivities), np.array(stabilities), np.array(metrics))
contrast_plot = vectorplots.contrast_streams(*args, vlines=contrasts)
contrast_plot.savefig(fname, **savefig_opts)
vectorplots.close([contrast_plot])
# Composing the whole figure together
svg_path = os.path.join(output_dir, '%s.svg' % name)
model_list = ['L','AL','GCL','GCAL']
current_model = model_list[[6,7,8,9].index(fig)]
mapping = dict([('[CONTRAST%d]' % i,c) for (i,c) in enumerate(contrasts)]
+ [('[TIME]', int(max_time)),
('[Model-Name]', current_model + " Model")])
layers = dict((name,name==current_model) for name in model_list)
return Display(compose.apply_template(template_path, svg_path,
mapping=mapping,
layers=layers,
size_factor=size_factor))
def fig06_09(template_dir,
build_dir,
fig,
contrasts=(10, 25, 100),
selectivity_norm=1.0,
input_seed=102,
size_factor=1.0,
kernel=analysis.gamma_metric,
show_raster=False):
name = 'fig0%d' % fig
if show_raster: return Display(snapshot=name, template_dir=template_dir)
template_path, output_dir = setup('fig06_09',
template_dir,
build_dir,
subdir=name)
# Get the files containing the relevant data
output_name = ('Fig0%d' % fig) if fig != 6 else 'Fig05_06'
pattern = 'jn13_figures/output/%s/*%s_*/*/*.npz' % (output_name,
output_name)
files = lancet.FilePattern('npz_file', pattern)
info = lancet.FileInfo(files, 'npz_file', lancet.NumpyFile())
dframe = info.dframe
# Some basic constants
max_time = max(dframe['time'])
input_seeds = set(dframe['input_seed'])
all_contrasts = set(dframe['contrast'])
all_times = set(dframe['time'])
savefig_opts = dict(transparent=True,
format='svg',
bbox_inches='tight',
pad_inches=0)
# Saving all the png/svg components for the three chosen contrasts (subplot D-F)
boolarr = np.array(dframe['time'] == max_time) & np.array(
dframe['input_seed'] == input_seed)
for contrast in contrasts:
fname = os.path.join(output_dir,
'%%s_%d_%d.%%s' % (contrast, max_time))
contrast_slice = np.array(dframe['contrast'] == contrast)
row = info.load(dframe[boolarr & contrast_slice]).iloc[0]
OR_analysis(row,
template_dir,
output_dir,
name='OR_c%s' % contrast,
selectivity=True)
# Collecting preferences/selectivities across time/seeds for map_development plots
selectivities, stabilities = [], []
for seed in input_seeds:
seed_slice = np.array(dframe['input_seed'] == seed)
df = info.load(dframe[contrast_slice & seed_slice])
# Ensuring the data is sorted by simulation time
sorted_rows = [
df[df['time'] == time].iloc[0] for time in sorted(df['time'])
]
prefs = [get_pref(row) for row in sorted_rows]
# Each selectivity is a whole map, these selectivities are averages across the map
mean_sels = [get_sel(row).mean() for row in sorted_rows]
selectivities.append(mean_sels)
stabilities.append(analysis.stability_index(prefs))
# Save the stability/selectivity plots
map_dev = vectorplots.map_development_streams(
np.array(stabilities),
np.array(selectivities),
selectivity_norm=selectivity_norm)
fname = os.path.join(output_dir, 'SI_%s.svg' % contrast)
map_dev.savefig(fname, **savefig_opts)
vectorplots.close([map_dev])
# Collect the selectivity, stability and map metric across contrasts and seeds
selectivities, stabilities, metrics = [], [], []
time_slice = np.array(dframe['time'] == max_time)
for seed in input_seeds:
seed_slice = np.array(dframe['input_seed'] == seed)
df = info.load(dframe[time_slice & seed_slice])
# Ensuring contrasts are sorted
sorted_rows = [
df[df['contrast'] == contrast].iloc[0]
for contrast in sorted(df['contrast'])
]
missing_contrasts = set(df['contrast']) ^ set(all_contrasts)
if missing_contrasts != set():
missing = ', '.join(str(el) for el in missing_contrasts)
print "Warning: Contrast values %s are missing for seed %d" % (
missing, seed)
# Each selectivity is a whole map, these selectivities are averages across the map
mean_sels = [get_sel(row).mean() for row in sorted_rows]
# Computing map quality based on the pinwheel density
pinwheel_counts = [row['pinwheels'].shape[0] for row in sorted_rows]
# Pinwheel density is pinwheel count / (kmax**2)
pinwheel_densities = [(pwcount / row['kmax']**2)
for (pwcount,
row) in zip(pinwheel_counts, sorted_rows)]
metrics.append([kernel(pwd) for pwd in pinwheel_densities])
selectivities.append(mean_sels)
# Obtaining the mean stability per contrast, sorted by contrast then by time
df = info.load(dframe[seed_slice])
contrast_stabilities = []
for contrast in sorted(all_contrasts):
rows = [
df[(df['time'] == time) & (df['contrast'] == contrast)].iloc[0]
for time in sorted(all_times)
]
prefs = [get_pref(row) for row in rows]
mean_stability = np.mean(analysis.stability_index(prefs))
contrast_stabilities.append(mean_stability)
stabilities.append(contrast_stabilities)
# Save the contrast plot
fname = os.path.join(output_dir, 'contrast.svg')
args = (sorted(all_contrasts), np.array(selectivities),
np.array(stabilities), np.array(metrics))
contrast_plot = vectorplots.contrast_streams(*args, vlines=contrasts)
contrast_plot.savefig(fname, **savefig_opts)
vectorplots.close([contrast_plot])
# Composing the whole figure together
svg_path = os.path.join(output_dir, '%s.svg' % name)
model_list = ['L', 'AL', 'GCL', 'GCAL']
current_model = model_list[[6, 7, 8, 9].index(fig)]
mapping = dict([('[CONTRAST%d]' % i, c)
for (i, c) in enumerate(contrasts)] +
[('[TIME]', int(max_time)),
('[Model-Name]', current_model + " Model")])
layers = dict((name, name == current_model) for name in model_list)
return Display(
compose.apply_template(template_path,
svg_path,
mapping=mapping,
layers=layers,
size_factor=size_factor))