def computePairWiseConfusions_old(objects_oi, OUTPATH=None):
""" For a set of objects and features, run classifiers,
and output trial structures for all 2x2 tasks.
"""
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)
all_features, all_metas = obj.getAllFeatures(objs_oi)
# features_oi = ['Caffe_fc6', 'Caffe_fc7', 'Caffe', 'VGG_fc6', 'VGG_fc7', 'VGG']
features_oi = ['Alexnet_fc6', 'Alexnet_fc7', 'Alexnet']
subsample = None
nsamples_noisemodel = 2
result = {}
for feat in features_oi:
if feat not in all_features.key():
continue
features = all_features[feat]
meta = all_metas[feat]
tasks = getBinaryTasks(meta, tasks_oi)
trials, trials_io = [], []
if 'rep' in feat:
noise_model = 'rep' #'rep'
else:
noise_model = None
print 'Running machine_objectome : ' + str(feat) + ': ' + str(features.shape) + '\t' + str(noise_model)
for isample in range(nsamples_noisemodel):
features_sample = sampleFeatures(features, noise_model, subsample)
for task in tasks:
p_, p_s_, t_, t_s_ = testFeatures_base(features, meta, task, objs_oi)
trials.extend(t_)
trials = format_trials_var(trials)
if OUTPATH != None:
if not os.path.exists(OUTPATH):
os.makedirs(OUTPATH)
if subsample != None:
feat = feat + '_' + str(subsample)
print 'Saved to ' + OUTPATH
save_trials(trials, objs_oi, OUTPATH + feat + 'full_var_bg.mat')
result[feat] = trials
return result
def computePairWiseConfusions(objects_oi, OUTPATH=None, IMGPATH=None):
""" For a set of objects and features, run classifiers,
and output trial structures for all 2x2 tasks.
"""
features_oi = ['PXLn', 'V1', 'HMAX', 'Alexnet_fc6', 'Alexnet_fc7', 'Alexnet', 'VGG_fc6', 'VGG_fc7', 'VGG', 'Googlenet', 'Resnet']
all_features, all_metas = obj.getAllFeatures(objects_oi, IMGPATH)
result = testFeatures(all_features, all_metas, features_oi, objects_oi)
for feat in features_oi:
if feat not in all_features.keys():
continue
if OUTPATH != None:
if not os.path.exists(OUTPATH):
os.makedirs(OUTPATH)
print 'Save to ' + OUTPATH
save_trials(result[feat], result['objs_oi'], OUTPATH + feat + 'full_var_bg.mat')
return
def computePairWiseConfusions_base(objects_oi, OUTPATH=None, silence_mode=0):
""" For a set of objects, compute pixel and v1 features, run classifiers,
and output trial structures for all 2x2 tasks.
"""
if type(objects_oi) is dict:
objs_oi = objects_oi['objs']
tasks_oi = objects_oi['tasks']
else:
tasks_oi = objects_oi
objs_oi = objects_oi
run_silencing_exp = silence_mode > 0
print 'Loading features...'
if run_silencing_exp:
all_features = pk.load(open('quickload_feature_data/features.pkl', 'r'))
all_metas = pk.load(open('quickload_feature_data/metas.pkl', 'r'))
else:
all_features, all_metas = obj.getAllFeatures(objs_oi)
features_oi = ['IT', 'V4']
result = {}
for feat in features_oi:
print 'Running machine_objectome : \n' + str(objs_oi) + '\n ' + str(feat) + '\n\n'
features = all_features[feat]
meta = all_metas[feat]
tasks = getBinaryTasks(meta, tasks_oi)
trials, trials_io = [], []
trials_s = {}
if run_silencing_exp:
features_s = silenceFeature(features, silence_mode)
else:
features_s = None
for task in tasks:
p_, p_s_, t_, t_io_, t_s_ = getPerformanceFromFeatures_base(features, meta, task, objs_oi, features_s, nsplits=50)
trials.extend(t_)
trials_io.extend(t_io_)
if run_silencing_exp:
for fs in features_s:
if not (fs in t_s_):
trials_s[fs] = []
trials_s[fs].extend(t_s_[fs])
trials = format_trials_var(trials)
trials_io = format_trials_var(trials_io)
if run_silencing_exp:
trials_s = format_trials_var(trials_s)
if OUTPATH != None:
ts = time.time()
st = datetime.datetime.fromtimestamp(ts).strftime('%M.%H.%d%m')
print 'Saving to ' + OUTPATH
if not os.path.exists(OUTPATH):
os.makedirs(OUTPATH)
if run_silencing_exp:
save_trials(trials, objs_oi, OUTPATH + feat + st + '.mat')
save_trials(trials_s, objs_oi, OUTPATH + feat + st + 'sil2.mat')
else:
save_trials(trials, objs_oi, OUTPATH + feat + 'full_var_bg.mat')
save_trials(trials_io, objs_oi, OUTPATH + feat + 'full_var_bg_ideal_obs.mat')
result[feat] = trials
return result