def composite_dataset(dataset="objectome24", threshold=12000, mongo_reload=False):
if dataset == "objectome24":
collections = ["objectome64", "objectome_imglvl", "ko_obj24_basic_2ways", "monkobjectome"]
meta = obj.objectome24_meta()
elif dataset == "hvm10":
collections = [
"hvm10_basic_2ways",
"hvm10_allvar_basic_2ways",
] # , 'hvm10_basic_2ways_newobj', 'hvm10-finegrain']
meta = obj.hvm10_meta()
fns = ["sample_obj", "id", "dist_obj", "choice", "WorkerID"]
col_data = ()
for col in collections:
dset = obj.psychophysDatasetObject(col, {}, meta, mongo_reload=mongo_reload)
col_data = col_data + (dset.trials,)
trials = tb.rowstack(col_data)
# segregate into pool and individuals
workers = trials["WorkerID"]
col_data_seg = {"all": trials, "pool": ()}
for uw in np.unique(workers):
tw = np.nonzero([w == uw for w in workers])[0]
if len(tw) < threshold:
col_data_seg["pool"] = col_data_seg["pool"] + (trials[tw],)
else:
col_data_seg[uw] = trials[tw]
col_data_seg["pool"] = tb.rowstack(col_data_seg["pool"])
return col_data_seg
def testFeatures(all_features, all_metas, features_oi, objects_oi):
if type(objects_oi) is dict:
objs_oi = objects_oi['objs']
tasks_oi = objects_oi['tasks']
else:
tasks_oi = np.array(objects_oi)
objs_oi = np.array(objects_oi)
subsample = 1000
noise_model = None
nsamples_noisemodel = 10
nsplits = 100
result = {'objs_oi':objs_oi}
task_trials = ()
rec = {}
for feat in features_oi:
if feat not in all_features.keys():
continue
features = all_features[feat]
meta = fix_meta(all_metas[feat])
tasks = getBinaryTasks(meta, tasks_oi)
print 'Running machine_objectome : ' + str(feat) + ': ' + str(features.shape)
for isample in range(nsamples_noisemodel):
features_sample = sampleFeatures(features, noise_model, subsample)
for task in tasks:
trials = testFeatures_base(features_sample, meta, task, objs_oi, nsplits=nsplits)
task_trials = task_trials + (trials,)
task_trials = tb.rowstack(task_trials)
task_trials['WorkerID'] = feat
rec[feat] = task_trials
return task_trials