def load_responses(response_file, stimuli):
neural_response_file = response_file + "neural_responses.npy"
neural_responses = np.load(neural_response_file)
brodmann_file = response_file + "brodmann_areas.npy"
brodmann_locations = np.load(brodmann_file)
assembly = xr.DataArray(
neural_responses,
coords={
'image_num':
('presentation', list(range(neural_responses.shape[0]))),
'image_id': ('presentation', [
stimuli['image_id'][stimuli['image_number'] == num].values[0]
for num in range(neural_responses.shape[0])
]),
'region': ('neuroid', brodmann_locations),
'neuroid_id': ('neuroid', list(range(neural_responses.shape[1]))),
'time': ('time_bin', np.linspace(0, 1, 32)),
'time_bin_start': ('time_bin', np.arange(0, 1000, 31.25)),
'time_bin_end': ('time_bin', np.arange(31.25, 1001, 31.25))
},
dims=['presentation', 'neuroid', 'time_bin'])
assembly = NeuronRecordingAssembly(assembly)
assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
return assembly
def load_responses(response_file, stimuli):
responses = h5py.File(response_file, 'r')
assemblies = []
neuroid_id_offset = 0
for monkey in responses.keys():
spike_rates = responses[monkey]['rates']
assembly = xr.DataArray(spike_rates.value,
coords={
'image_num': ('image_id', list(range(spike_rates.shape[0]))),
'image_id': ('image_id', [
stimuli['image_id'][stimuli['image_number'] == num].values[0]
for num in range(spike_rates.shape[0])]),
'neuroid_id': ('neuroid', list(
range(neuroid_id_offset, neuroid_id_offset + spike_rates.shape[1]))),
'region': ('neuroid', ['IT'] * spike_rates.shape[1]),
'monkey': ('neuroid', [monkey] * spike_rates.shape[1]),
'repetition': list(range(spike_rates.shape[2])),
},
dims=['image_id', 'neuroid', 'repetition'])
assemblies.append(assembly)
neuroid_id_offset += spike_rates.shape[1]
assembly = xr.concat(assemblies, 'neuroid')
assembly = assembly.stack(presentation=['image_id', 'repetition'])
assembly = NeuronRecordingAssembly(assembly)
assert len(assembly['presentation']) == 1600 * 45
assert len(np.unique(assembly['image_id'])) == 1600
assert len(assembly.sel(monkey='nano')['neuroid']) == len(assembly.sel(monkey='magneto')['neuroid']) == 288
assert len(assembly['neuroid']) == len(np.unique(assembly['neuroid_id'])) == 288 * 2
return assembly
def collect_synth(h5, data_dir):
protos_stimuli = []
responses_synth_d = {}
for monkey in h5.root.images.synthetic:
for setting in monkey:
for session_images in setting:
session_neural = h5.root.neural.synthetic[monkey._v_name][
setting._v_name][session_images._v_name]
session_target_inds = h5.root.target_inds[monkey._v_name][
setting._v_name][session_images._v_name]
identifier = f"{monkey._v_name[-1]}_{setting._v_name}_{session_images._v_name}"
img_temp_path = data_dir / "images_temp" / "synthetic" / identifier
img_temp_path.mkdir(parents=True, exist_ok=True)
proto_stimuli = np_to_png(session_images, img_temp_path)
proto_stimuli["animal"] = monkey._v_name
proto_stimuli["setting"] = setting._v_name
proto_stimuli["session"] = session_images._v_name
protos_stimuli.append(proto_stimuli)
proto_neural = np_to_xr(monkey, setting, session_neural,
proto_stimuli, session_target_inds,
"synth")
proto_neural = NeuronRecordingAssembly(proto_neural)
responses_synth_d[proto_neural.name] = proto_neural
proto_stimuli_all = pd.concat(protos_stimuli, axis=0)
assert len(np.unique(
proto_stimuli_all['image_id'])) == len(proto_stimuli_all)
stimuli = StimulusSet(proto_stimuli_all)
stimuli.image_paths = {
row.image_id: row.image_current_local_file_path
for row in stimuli.itertuples()
}
return stimuli, responses_synth_d
def load_responses(data_dir, stimuli):
IT_base616 = pickle.load(
open(os.path.join(data_dir, 'data_IT_base616.pkl'), 'rb'))
features = IT_base616[
'IT_features'] # Shaped images x neuroids x repetitions x time_bins
# Drop all time_bins except the fifth, which corresponds to 70-170ms
# For future reference the time-bins are as follows:
# 70-120ms, 120-170ms, 170-220ms, 220-270ms, 70-170ms, 170-270ms, 70-270ms
features = features[:, :, :, 4]
features = features[:, :, :, np.newaxis]
# Drop all repetitions beyond 33rd (all neuroids have at least 27, but none have greater than 33)
features = features[:, :, :33, :]
neuroid_meta = pickle._Unpickler(
open(os.path.join(data_dir, 'IT_neural_meta_full.pkl'), 'rb'))
neuroid_meta.encoding = 'latin1'
neuroid_meta = neuroid_meta.load()
assembly = xr.DataArray(
features,
coords={
'region': ('neuroid', ['IT'] * len(neuroid_meta)),
'neuroid_id': ('neuroid', list(range(features.shape[1]))),
'time_bin_start': ('time_bin', [70]),
'time_bin_stop': ('time_bin', [170]),
'repetition': ('repetition', list(range(features.shape[2])))
},
dims=['image', 'neuroid', 'repetition', 'time_bin'])
for column_name, column_data in neuroid_meta.iteritems():
assembly = assembly.assign_coords(
**{f'{column_name}': ('neuroid', list(column_data.values))})
for column_name, column_data in stimuli.iteritems():
assembly = assembly.assign_coords(
**{f'{column_name}': ('image', list(column_data.values))})
# Collapse dimensions 'image' and 'repetitions' into a single 'presentation' dimension
assembly = assembly.stack(
presentation=('image', 'repetition')).reset_index('presentation')
assembly = assembly.drop('image')
assembly = NeuronRecordingAssembly(assembly)
assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
return assembly
def collect_responses_nat(h5, stimuli):
responses_nat_d = {}
for monkey in h5.root.neural.naturalistic:
for setting in monkey:
for session in setting:
target_inds_session = h5.root.target_inds[monkey._v_name][
setting._v_name][session._v_name]
proto = np_to_xr(monkey, setting, session, stimuli,
target_inds_session, "nat")
proto = NeuronRecordingAssembly(proto)
responses_nat_d[proto.name] = proto
return responses_nat_d
def load_responses(response_file, stimuli):
responses = h5py.File(response_file, 'r')
assemblies = []
neuroid_id_offset = 0
for monkey in responses.keys():
spike_rates = responses[monkey]['rates']
assembly = xr.DataArray(spike_rates.value,
coords={
'image_num': ('image_id', list(range(spike_rates.shape[0]))),
'image_id': ('image_id', [
stimuli['image_id'][stimuli['image_number'] == num].values[0]
for num in range(spike_rates.shape[0])]),
'neuroid_id': ('neuroid', list(
range(neuroid_id_offset, neuroid_id_offset + spike_rates.shape[1]))),
'region': ('neuroid', ['IT'] * spike_rates.shape[1]),
'monkey': ('neuroid', [monkey] * spike_rates.shape[1]),
'repetition': list(range(spike_rates.shape[2])),
},
dims=['image_id', 'neuroid', 'repetition'])
assemblies.append(assembly)
neuroid_id_offset += spike_rates.shape[1]
assembly = xr.concat(assemblies, 'neuroid')
assembly = assembly.stack(presentation=['image_id', 'repetition'])
assembly = NeuronRecordingAssembly(assembly)
assert len(assembly['presentation']) == 1600 * 45
assert len(np.unique(assembly['image_id'])) == 1600
assert len(assembly.sel(monkey='nano')['neuroid']) == len(assembly.sel(monkey='magneto')['neuroid']) == 288
assert len(assembly['neuroid']) == len(np.unique(assembly['neuroid_id'])) == 288 * 2
# filter noisy electrodes
assembly = filter_neuroids(assembly, threshold=.7)
# add time info
assembly = assembly.expand_dims('time_bin')
assembly['time_bin_start'] = 'time_bin', [70]
assembly['time_bin_end'] = 'time_bin', [170]
assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
return assembly
def load_responses(data_dir, stimuli):
data_dir = data_dir / 'database'
assert os.path.isdir(data_dir)
psth = np.load(data_dir / 'solo.rsvp.things-2.experiment_psth.npy') # Shaped images x repetitions x time_bins x channels
# Compute firing rate for given time bins
timebins = [[70, 170], [170, 270], [50, 100], [100, 150], [150, 200], [200, 250], [70, 270]]
photodiode_delay = 30 # Delay recorded on photodiode is ~30ms
timebase = np.arange(-100, 381, 10) # PSTH from -100ms to 380ms relative to stimulus onset
assert len(timebase) == psth.shape[2]
rate = np.empty((len(timebins), psth.shape[0], psth.shape[1], psth.shape[3]))
for idx, tb in enumerate(timebins):
t_cols = np.where((timebase >= (tb[0] + photodiode_delay)) & (timebase < (tb[1] + photodiode_delay)))[0]
rate[idx] = np.mean(psth[:, :, t_cols, :], axis=2) # Shaped time bins x images x repetitions x channels
assembly = xr.DataArray(rate,
coords={'repetition': ('repetition', list(range(rate.shape[2]))),
'time_bin_id': ('time_bin', list(range(rate.shape[0]))),
'time_bin_start': ('time_bin', [x[0] for x in timebins]),
'time_bin_stop': ('time_bin', [x[1] for x in timebins])},
dims=['time_bin', 'image', 'repetition', 'neuroid'])
# Add neuroid related meta data
neuroid_meta = pd.DataFrame(json.load(open(data_dir.parent / 'array-metadata' / 'mapping.json')))
for column_name, column_data in neuroid_meta.iteritems():
assembly = assembly.assign_coords(**{f'{column_name}': ('neuroid', list(column_data.values))})
# Add stimulus related meta data
for column_name, column_data in stimuli.iteritems():
assembly = assembly.assign_coords(**{f'{column_name}': ('image', list(column_data.values))})
# Collapse dimensions 'image' and 'repetitions' into a single 'presentation' dimension
assembly = assembly.stack(presentation=('image', 'repetition')).reset_index('presentation')
assembly = assembly.drop('image')
assembly = NeuronRecordingAssembly(assembly)
# Filter noisy electrodes
psth = np.load(data_dir / 'solo.rsvp.things-2.normalizer_psth.npy')
t_cols = np.where((timebase >= (70 + photodiode_delay)) & (timebase < (170 + photodiode_delay)))[0]
rate = np.mean(psth[:, :, t_cols, :], axis=2)
normalizer_assembly = xr.DataArray(rate,
coords={'repetition': ('repetition', list(range(rate.shape[1]))),
'image_id': ('image', list(range(rate.shape[0]))),
'id': ('image', list(range(rate.shape[0])))},
dims=['image', 'repetition', 'neuroid'])
for column_name, column_data in neuroid_meta.iteritems():
normalizer_assembly = normalizer_assembly.assign_coords(
**{f'{column_name}': ('neuroid', list(column_data.values))})
normalizer_assembly = normalizer_assembly.stack(presentation=('image', 'repetition')).reset_index('presentation')
normalizer_assembly = normalizer_assembly.drop('image')
normalizer_assembly = normalizer_assembly.transpose('presentation', 'neuroid')
normalizer_assembly = NeuronRecordingAssembly(normalizer_assembly)
filtered_assembly = filter_neuroids(normalizer_assembly, 0.7)
assembly = assembly.sel(neuroid=np.isin(assembly.neuroid_id, filtered_assembly.neuroid_id))
assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
# Add other experiment and data processing related info
assembly.attrs['image_size_degree'] = 8
assembly.attrs['stim_on_time_ms'] = 100
return assembly
def load_responses(data_dir, stimuli):
psth = np.load(data_dir / 'solo.rsvp.hvm.experiment_psth.npy'
) # Shaped images x repetitions x time_bins x channels
# Drop first (index 0) and second last session (index 25) since they had only one repetition each
# Actually not, since we're sticking to older protocol re: data cleaning for now
# psth = np.delete(psth, (0, 25), axis=1)
# Compute firing rate for given time bins
timebins = [[70, 170], [170, 270], [50, 100], [100, 150], [150, 200],
[200, 250], [70, 270]]
photodiode_delay = 30 # Delay recorded on photodiode is ~30ms
timebase = np.arange(
-100, 381, 10) # PSTH from -100ms to 380ms relative to stimulus onset
assert len(timebase) == psth.shape[2]
rate = np.empty(
(len(timebins), psth.shape[0], psth.shape[1], psth.shape[3]))
for idx, tb in enumerate(timebins):
t_cols = np.where((timebase >= (tb[0] + photodiode_delay))
& (timebase < (tb[1] + photodiode_delay)))[0]
rate[idx] = np.mean(
psth[:, :, t_cols, :],
axis=2) # Shaped time bins x images x repetitions x channels
# Load image related meta data (id ordering differs from dicarlo.hvm)
image_id = [
x.split()[0][:-4] for x in open(data_dir.parent / 'image-metadata' /
'hvm_map.txt').readlines()
]
# Load neuroid related meta data
neuroid_meta = pd.DataFrame(
json.load(open(data_dir.parent / 'array-metadata' / 'mapping.json')))
assembly = xr.DataArray(
rate,
coords={
'repetition': ('repetition', list(range(rate.shape[2]))),
'time_bin_id': ('time_bin', list(range(rate.shape[0]))),
'time_bin_start': ('time_bin', [x[0] for x in timebins]),
'time_bin_stop': ('time_bin', [x[1] for x in timebins]),
'image_id': ('image', image_id)
},
dims=['time_bin', 'image', 'repetition', 'neuroid'])
for column_name, column_data in neuroid_meta.iteritems():
assembly = assembly.assign_coords(
**{f'{column_name}': ('neuroid', list(column_data.values))})
assembly = assembly.sortby(assembly.image_id)
stimuli = stimuli.sort_values(by='image_id').reset_index(drop=True)
for column_name, column_data in stimuli.iteritems():
assembly = assembly.assign_coords(
**{f'{column_name}': ('image', list(column_data.values))})
assembly = assembly.sortby(
assembly.id) # Re-order by id to match dicarlo.hvm ordering
# Collapse dimensions 'image' and 'repetitions' into a single 'presentation' dimension
assembly = assembly.stack(
presentation=('image', 'repetition')).reset_index('presentation')
assembly = NeuronRecordingAssembly(assembly)
# Filter noisy electrodes
psth = np.load(data_dir / 'solo.rsvp.hvm.normalizer_psth.npy')
t_cols = np.where((timebase >= (70 + photodiode_delay))
& (timebase < (170 + photodiode_delay)))[0]
rate = np.mean(psth[:, :, t_cols, :], axis=2)
normalizer_assembly = xr.DataArray(
rate,
coords={
'repetition': ('repetition', list(range(rate.shape[1]))),
'image_id': ('image', list(range(rate.shape[0]))),
'id': ('image', list(range(rate.shape[0])))
},
dims=['image', 'repetition', 'neuroid'])
for column_name, column_data in neuroid_meta.iteritems():
normalizer_assembly = normalizer_assembly.assign_coords(
**{f'{column_name}': ('neuroid', list(column_data.values))})
normalizer_assembly = normalizer_assembly.stack(
presentation=('image', 'repetition')).reset_index('presentation')
normalizer_assembly = normalizer_assembly.drop('image')
normalizer_assembly = normalizer_assembly.transpose(
'presentation', 'neuroid')
normalizer_assembly = NeuronRecordingAssembly(normalizer_assembly)
filtered_assembly = filter_neuroids(normalizer_assembly, 0.7)
assembly = assembly.sel(
neuroid=np.isin(assembly.neuroid_id, filtered_assembly.neuroid_id))
assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
# Add other experiment related info
assembly.attrs['image_size_degree'] = 8
assembly.attrs['stim_on_time_ms'] = 100
return assembly