def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('dataset', type=str)
parser.add_argument('output', type=str, nargs='+')
parser.add_argument('--model', '-m', type=str, default='')
parser.add_argument(
'--preprocess',
'-p',
type=int,
default=0,
help=
'If you haven\'t already applied `preprocess` to the data, set to 1 (default: 0).'
)
parser.add_argument('--verbosity', '-v', type=int, default=1)
args = parser.parse_args(argv[1:])
experiment = Experiment()
# load data
data = load_data(args.dataset)
if args.preprocess:
# preprocess data
data = preprocess(data, args.verbosity)
if args.model:
# load training results
results = Experiment(args.model)['models']
else:
# use default model
results = None
# predict firing rates
data = predict(data, results, verbosity=args.verbosity)
# remove data except predictions
for entry in data:
if 'spikes' in entry:
del entry['spikes']
if 'spike_times' in entry:
del entry['spike_times']
del entry['calcium']
for filepath in args.output:
if filepath.lower().endswith('.mat'):
# store in MATLAB format
savemat(filepath, {'data': data})
else:
with open(filepath, 'w') as handle:
dump(data, handle, protocol=2)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('dataset', type=str)
parser.add_argument('predictions', type=str, nargs='?')
parser.add_argument('--downsampling', '-s', type=int, default=[1, 2, 3, 4, 5, 10, 15, 20, 25, 30, 40, 50], nargs='+')
parser.add_argument('--optimize', '-z', type=int, default=1,
help='Whether or not to optimize point-wise nonlinearity when evaluating likelihood.')
parser.add_argument('--regularization', '-r', type=float, default=5e-8,
help='Controls smoothness of optimized nonlinearity (default: 5e-8).')
parser.add_argument('--method', '-m', type=str, default='corr', choices=['corr', 'auc', 'info'])
parser.add_argument('--weighted-average','-w', type=int, default=0,
help='Whether or not traces to weight traces by their duration.')
parser.add_argument('--output', '-o', type=str, default='')
parser.add_argument('--verbosity', '-v', type=int, default=1)
args, _ = parser.parse_known_args(argv[1:])
experiment = Experiment()
data = load_data(args.dataset)
if not args.predictions:
# use raw calcium signal for prediction
calcium_min = min(hstack(entry['calcium'] for entry in data))
for entry in data:
entry['predictions'] = entry['calcium'] - calcium_min + 1e-5
else:
predictions = load_data(args.predictions)
try:
if len(predictions) != len(data):
raise ValueError()
for entry1, entry2 in zip(data, predictions):
if entry1['calcium'].size != entry2['predictions'].size:
raise ValueError()
entry1['predictions'] = entry2['predictions']
except ValueError:
print 'These predictions seem to be for a different dataset.'
return 1
fps = []
loglik = []
correlations = []
auc = []
entropy = []
functions = []
for ds in args.downsampling:
if args.verbosity > 0:
if args.weighted_average:
if args.method.lower().startswith('c'):
print '{0:>5} {1:>7} {2:>7} {3}'.format('Cell', '#Traces', 'FPS ', 'Correlation')
elif args.method.lower().startswith('a'):
print '{0:>5} {1:>7} {2:>7} {3}'.format('Cell', '#Traces', 'FPS ', 'AUC')
else:
print '{0:>5} {1:>7} {2:>7} {3}'.format('Cell', '#Traces', 'FPS ', 'Information gain')
else:
if args.method.lower().startswith('c'):
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ', 'Correlation')
elif args.method.lower().startswith('a'):
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ', 'AUC')
else:
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ', 'Information gain')
fps.append([])
for entry in data:
fps[-1].append(entry['fps'] / ds)
if args.method.lower().startswith('c'):
# compute correlations
R = evaluate(data, method=args.method,
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity)
correlations.append(R)
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(R, data, ds)
else:
print_traces(R, fps)
elif args.method.lower().startswith('a'):
# compute correlations
A = evaluate(data, method=args.method,
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity)
auc.append(A)
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(A, data, ds)
else:
print_traces(A, fps)
else:
# compute log-likelihoods
L, H, f = evaluate(data, method='loglik',
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity,
return_all=True,
regularize=args.regularization)
loglik.append(L)
entropy.append(H)
functions.append((f.x, f.y))
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(H + L, data, ds)
else:
print_traces(H + L, fps)
if args.output.lower().endswith('.mat'):
if args.method.lower().startswith('c'):
savemat(args.output, {'fps': asarray(fps), 'correlations': asarray(correlations)})
elif args.method.lower().startswith('a'):
savemat(args.output, {'fps': asarray(fps), 'auc': asarray(auc)})
else:
savemat(args.output, {
'fps': asarray(fps),
'loglik': asarray(loglik),
'entropy': asarray(entropy),
'info': asarray(loglik) + asarray(entropy)})
elif args.output:
if os.path.isdir(args.output):
filepath = os.path.join(args.output, args.method + '.{0}.{1}.xpck')
else:
filepath = args.output
experiment['args'] = args
experiment['fps'] = asarray(fps)
if args.method.lower().startswith('c'):
experiment['correlations'] = asarray(correlations)
elif args.method.lower().startswith('a'):
experiment['auc'] = asarray(auc)
else:
experiment['loglik'] = asarray(loglik)
experiment['entropy'] = asarray(entropy)
experiment['info'] = asarray(loglik) + asarray(entropy)
experiment['f'] = functions
experiment.save(filepath, overwrite=True)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('dataset', type=str, nargs='+')
parser.add_argument('output', type=str)
parser.add_argument('--num_components', '-c', type=int, default=3)
parser.add_argument('--num_features', '-f', type=int, default=2)
parser.add_argument('--num_models', '-m', type=int, default=4)
parser.add_argument('--keep_all', '-k', type=int, default=1)
parser.add_argument('--finetune', '-n', type=int, default=0)
parser.add_argument('--num_valid', '-s', type=int, default=0)
parser.add_argument('--var_explained', '-e', type=float, default=95.)
parser.add_argument('--window_length', '-w', type=float, default=1000.)
parser.add_argument('--regularize', '-r', type=float, default=0.)
parser.add_argument('--preprocess', '-p', type=int, default=0)
parser.add_argument('--verbosity', '-v', type=int, default=1)
args, _ = parser.parse_known_args(argv[1:])
experiment = Experiment()
# load data
data = []
for dataset in args.dataset:
data = data + load_data(dataset)
# preprocess data
if args.preprocess:
data = preprocess(data)
# list of all cells
if 'cell_num' in data[0]:
# several trials/entries may belong to the same cell
cells = unique([entry['cell_num'] for entry in data])
else:
# one cell corresponds to one trial/entry
cells = range(len(data))
for i in cells:
data[i]['cell_num'] = i
for i in cells:
data_train = [entry for entry in data if entry['cell_num'] != i]
data_test = [entry for entry in data if entry['cell_num'] == i]
if args.verbosity > 0:
print 'Test cell: {0}'.format(i)
# train on all cells but cell i
results = train(
data=data_train,
num_valid=args.num_valid,
num_models=args.num_models,
var_explained=args.var_explained,
window_length=args.window_length,
keep_all=args.keep_all,
finetune=args.finetune,
model_parameters={
'num_components': args.num_components,
'num_features': args.num_features},
training_parameters={
'verbosity': 0},
regularize=args.regularize,
verbosity=1)
if args.verbosity > 0:
print 'Predicting...'
# predict responses of cell i
predictions = predict(data_test, results, verbosity=0)
for entry1, entry2 in zip(data_test, predictions):
entry1['predictions'] = entry2['predictions']
# remove data except predictions
for entry in data:
if 'spikes' in entry:
del entry['spikes']
if 'spike_times' in entry:
del entry['spike_times']
del entry['calcium']
# save results
if args.output.lower().endswith('.mat'):
savemat(args.output, convert({'data': data}))
elif args.output.lower().endswith('.xpck'):
experiment['args'] = args
experiment['data'] = data
experiment.save(args.output)
else:
with open(args.output, 'w') as handle:
dump(data, handle, protocol=2)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('dataset', type=str, nargs='+')
parser.add_argument('output', type=str)
parser.add_argument('--num_components', '-c', type=int, default=3)
parser.add_argument('--num_features', '-f', type=int, default=2)
parser.add_argument('--num_models', '-m', type=int, default=4)
parser.add_argument('--keep_all', '-k', type=int, default=1)
parser.add_argument('--finetune', '-n', type=int, default=0)
parser.add_argument('--num_valid', '-s', type=int, default=0)
parser.add_argument('--var_explained', '-e', type=float, default=95.)
parser.add_argument('--window_length', '-w', type=float, default=1000.)
parser.add_argument('--regularize', '-r', type=float, default=0.)
parser.add_argument('--preprocess', '-p', type=int, default=0)
parser.add_argument('--verbosity', '-v', type=int, default=1)
args, _ = parser.parse_known_args(argv[1:])
experiment = Experiment()
# load data
data = []
for dataset in args.dataset:
data = data + load_data(dataset)
# preprocess data
if args.preprocess:
data = preprocess(data)
# list of all cells
if 'cell_num' in data[0]:
# several trials/entries may belong to the same cell
cells = unique([entry['cell_num'] for entry in data])
else:
# one cell corresponds to one trial/entry
cells = range(len(data))
for i in cells:
data[i]['cell_num'] = i
for i in cells:
data_train = [entry for entry in data if entry['cell_num'] != i]
data_test = [entry for entry in data if entry['cell_num'] == i]
if args.verbosity > 0:
print 'Test cell: {0}'.format(i)
# train on all cells but cell i
results = train(data=data_train,
num_valid=args.num_valid,
num_models=args.num_models,
var_explained=args.var_explained,
window_length=args.window_length,
keep_all=args.keep_all,
finetune=args.finetune,
model_parameters={
'num_components': args.num_components,
'num_features': args.num_features
},
training_parameters={'verbosity': 0},
regularize=args.regularize,
verbosity=1)
if args.verbosity > 0:
print 'Predicting...'
# predict responses of cell i
predictions = predict(data_test, results, verbosity=0)
for entry1, entry2 in zip(data_test, predictions):
entry1['predictions'] = entry2['predictions']
# remove data except predictions
for entry in data:
if 'spikes' in entry:
del entry['spikes']
if 'spike_times' in entry:
del entry['spike_times']
del entry['calcium']
# save results
if args.output.lower().endswith('.mat'):
savemat(args.output, {'data': data})
elif args.output.lower().endswith('.xpck'):
experiment['args'] = args
experiment['data'] = data
experiment.save(args.output)
else:
with open(args.output, 'w') as handle:
dump(data, handle, protocol=2)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__,
formatter_class=lambda prog: HelpFormatter(prog, max_help_position=10, width=120))
parser.add_argument('dataset', type=str, nargs='+',
help='Dataset(s) used for training.')
parser.add_argument('output', type=str,
help='Directory or file where trained models will be stored.')
parser.add_argument('--num_components', '-c', type=int, default=3,
help='Number of components used in STM model (default: %(default)d).')
parser.add_argument('--num_features', '-f', type=int, default=2,
help='Number of quadratic features used in STM model (default: %(default)d).')
parser.add_argument('--num_models', '-m', type=int, default=4,
help='Number of models trained (predictions will be averaged across models, default: %(default)d).')
parser.add_argument('--keep_all', '-k', type=int, default=1,
help='If set to 0, only the best model of all trained models is kept (default: %(default)d).')
parser.add_argument('--finetune', '-n', type=int, default=0,
help='If set to 1, enables another finetuning step which is performed after training (default: %(default)d).')
parser.add_argument('--num_train', '-t', type=int, default=0,
help='If specified, a (random) subset of cells is used for training.')
parser.add_argument('--num_valid', '-s', type=int, default=0,
help='If specified, a (random) subset of cells will be used for early stopping based on validation error.')
parser.add_argument('--var_explained', '-e', type=float, default=95.,
help='Controls the degree of dimensionality reduction of fluorescence windows (default: %(default).0f).')
parser.add_argument('--window_length', '-w', type=float, default=1000.,
help='Length of windows extracted from calcium signal for prediction (in milliseconds, default: %(default).0f).')
parser.add_argument('--regularize', '-r', type=float, default=0.,
help='Amount of parameter regularization (filters are regularized for smoothness, default: %(default).1f).')
parser.add_argument('--preprocess', '-p', type=int, default=0,
help='If the data is not already preprocessed, this can be used to do it.')
parser.add_argument('--verbosity', '-v', type=int, default=1)
args, _ = parser.parse_known_args(argv[1:])
experiment = Experiment()
if not args.dataset:
print 'You have to specify at least 1 dataset.'
return 0
data = []
for dataset in args.dataset:
with open(dataset) as handle:
data = data + load(handle)
if args.preprocess:
data = preprocess(data, args.verbosity)
if 'cell_num' not in data[0]:
# no cell number is given, assume traces correspond to cells
for k, entry in enumerate(data):
entry['cell_num'] = k
# collect cell ids
cell_ids = unique([entry['cell_num'] for entry in data])
# pick cells for training
if args.num_train > 0:
training_cells = random_select(args.num_train, len(cell_ids))
else:
# use all cells for training
training_cells = range(len(cell_ids))
models = train([entry for entry in data if entry['cell_num'] in training_cells],
num_valid=args.num_valid,
num_models=args.num_models,
var_explained=args.var_explained,
window_length=args.window_length,
keep_all=args.keep_all,
finetune=args.finetune,
model_parameters={
'num_components': args.num_components,
'num_features': args.num_features},
training_parameters={
'verbosity': 1},
regularize=args.regularize,
verbosity=args.verbosity)
experiment['args'] = args
experiment['training_cells'] = training_cells
experiment['models'] = models
if os.path.isdir(args.output):
experiment.save(os.path.join(args.output, 'model.xpck'))
else:
experiment.save(args.output)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('dataset', type=str)
parser.add_argument('predictions', type=str, nargs='?')
parser.add_argument('--downsampling',
'-s',
type=int,
default=[1, 2, 3, 4, 5, 10, 15, 20, 25, 30, 40, 50],
nargs='+')
parser.add_argument(
'--optimize',
'-z',
type=int,
default=1,
help=
'Whether or not to optimize point-wise nonlinearity when evaluating likelihood.'
)
parser.add_argument(
'--regularization',
'-r',
type=float,
default=5e-8,
help='Controls smoothness of optimized nonlinearity (default: 5e-8).')
parser.add_argument('--method',
'-m',
type=str,
default='corr',
choices=['corr', 'auc', 'info'])
parser.add_argument(
'--weighted-average',
'-w',
type=int,
default=0,
help='Whether or not traces to weight traces by their duration.')
parser.add_argument('--output', '-o', type=str, default='')
parser.add_argument('--verbosity', '-v', type=int, default=1)
args, _ = parser.parse_known_args(argv[1:])
experiment = Experiment()
data = load_data(args.dataset)
if not args.predictions:
# use raw calcium signal for prediction
calcium_min = min(hstack(entry['calcium'] for entry in data))
for entry in data:
entry['predictions'] = entry['calcium'] - calcium_min + 1e-5
else:
predictions = load_data(args.predictions)
try:
if len(predictions) != len(data):
raise ValueError()
for entry1, entry2 in zip(data, predictions):
if entry1['calcium'].size != entry2['predictions'].size:
raise ValueError()
entry1['predictions'] = entry2['predictions']
except ValueError:
print 'These predictions seem to be for a different dataset.'
return 1
fps = []
loglik = []
correlations = []
auc = []
entropy = []
functions = []
for ds in args.downsampling:
if args.verbosity > 0:
if args.weighted_average:
if args.method.lower().startswith('c'):
print '{0:>5} {1:>7} {2:>7} {3}'.format(
'Cell', '#Traces', 'FPS ', 'Correlation')
elif args.method.lower().startswith('a'):
print '{0:>5} {1:>7} {2:>7} {3}'.format(
'Cell', '#Traces', 'FPS ', 'AUC')
else:
print '{0:>5} {1:>7} {2:>7} {3}'.format(
'Cell', '#Traces', 'FPS ', 'Information gain')
else:
if args.method.lower().startswith('c'):
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ',
'Correlation')
elif args.method.lower().startswith('a'):
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ', 'AUC')
else:
print '{0:>5} {1:>7} {2}'.format('Trace', 'FPS ',
'Information gain')
fps.append([])
for entry in data:
fps[-1].append(entry['fps'] / ds)
if args.method.lower().startswith('c'):
# compute correlations
R = evaluate(data,
method=args.method,
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity)
correlations.append(R)
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(R, data, ds)
else:
print_traces(R, fps)
elif args.method.lower().startswith('a'):
# compute correlations
A = evaluate(data,
method=args.method,
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity)
auc.append(A)
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(A, data, ds)
else:
print_traces(A, fps)
else:
# compute log-likelihoods
L, H, f = evaluate(data,
method='loglik',
optimize=args.optimize,
downsampling=ds,
verbosity=args.verbosity,
return_all=True,
regularize=args.regularization)
loglik.append(L)
entropy.append(H)
functions.append((f.x, f.y))
if args.verbosity > 0:
if args.weighted_average:
print_weighted_average(H + L, data, ds)
else:
print_traces(H + L, fps)
if args.output.lower().endswith('.mat'):
if args.method.lower().startswith('c'):
savemat(
args.output,
convert({
'fps': asarray(fps),
'correlations': asarray(correlations)
}))
elif args.method.lower().startswith('a'):
savemat(args.output,
convert({
'fps': asarray(fps),
'auc': asarray(auc)
}))
else:
savemat(
args.output,
convert({
'fps': asarray(fps),
'loglik': asarray(loglik),
'entropy': asarray(entropy),
'info': asarray(loglik) + asarray(entropy)
}))
elif args.output:
if os.path.isdir(args.output):
filepath = os.path.join(args.output, args.method + '.{0}.{1}.xpck')
else:
filepath = args.output
experiment['args'] = args
experiment['fps'] = asarray(fps)
if args.method.lower().startswith('c'):
experiment['correlations'] = asarray(correlations)
elif args.method.lower().startswith('a'):
experiment['auc'] = asarray(auc)
else:
experiment['loglik'] = asarray(loglik)
experiment['entropy'] = asarray(entropy)
experiment['info'] = asarray(loglik) + asarray(entropy)
experiment['f'] = functions
experiment.save(filepath, overwrite=True)
return 0