def predict_activitiy(index,fold,checkpoint):
path_to_dataset = '/scr/ashesh/activity-anticipation/dataset/{0}'.format(fold)
path_to_checkpoints = '/scr/ashesh/activity-anticipation/checkpoints/{0}'.format(fold)
test_data = cPickle.load(open('{1}/test_data_{0}.pik'.format(index,path_to_dataset)))
Y_te = test_data['labels']
X_te = test_data['features']
ground_truth_test_data = cPickle.load(open('{1}/ground_truth_test_data_{0}.pik'.format(index,path_to_dataset)))
Y_te_ground_truth = ground_truth_test_data['labels']
rnn = load('{2}/{0}/checkpoint.{1}'.format(index,checkpoint,path_to_checkpoints))
predictions = []
errors = 0
N = 0
for xte,yte,gte in zip(X_te,Y_te,Y_te_ground_truth):
prediction = rnn.predict_output(xte,OutputMaxProb)
predictions.append(prediction)
t = np.nonzero(yte-prediction)
'''
print "{0}".format(gte[:,0])
print "{0}".format(yte[:,0])
print "{0}".format(prediction[:,0])
print ""
#print t
'''
errors += len(t[0])
N += yte.shape[0]
#cPickle.dump(predictions,open('{1}/prediction_{0}.pik'.format(index,path_to_dataset),'wb'))
print 'error = {0}'.format(errors*1.0/N)
return (errors*1.0/N)
def predict_activitiy(index, fold, checkpoint):
path_to_dataset = '/scr/ashesh/activity-anticipation/dataset/{0}'.format(
fold)
path_to_checkpoints = '/scr/ashesh/activity-anticipation/checkpoints/{0}'.format(
fold)
test_data = cPickle.load(
open('{1}/test_data_{0}.pik'.format(index, path_to_dataset)))
Y_te = test_data['labels']
X_te = test_data['features']
ground_truth_test_data = cPickle.load(
open('{1}/ground_truth_test_data_{0}.pik'.format(
index, path_to_dataset)))
Y_te_ground_truth = ground_truth_test_data['labels']
rnn = load('{2}/{0}/checkpoint.{1}'.format(index, checkpoint,
path_to_checkpoints))
predictions = []
errors = 0
N = 0
for xte, yte, gte in zip(X_te, Y_te, Y_te_ground_truth):
prediction = rnn.predict_output(xte, OutputMaxProb)
predictions.append(prediction)
t = np.nonzero(yte - prediction)
'''
print "{0}".format(gte[:,0])
print "{0}".format(yte[:,0])
print "{0}".format(prediction[:,0])
print ""
#print t
'''
errors += len(t[0])
N += yte.shape[0]
#cPickle.dump(predictions,open('{1}/prediction_{0}.pik'.format(index,path_to_dataset),'wb'))
print 'error = {0}'.format(errors * 1.0 / N)
return (errors * 1.0 / N)
def evaluate(index,fold,checkpoint,model_type='lstm_one_layer',path_to_load_from=''):
path_to_dataset = '/scr/ashesh/brain4cars/dataset/{0}'.format(fold)
if len(path_to_load_from) > 0:
path_to_dataset = path_to_load_from
path_to_checkpoints = '/scr/ashesh/brain4cars/checkpoints/{0}'.format(fold)
test_data = cPickle.load(open('{1}/test_data_{0}.pik'.format(index,path_to_dataset)))
Y_te = test_data['labels']
X_te = test_data['features']
actions = []
if test_data.has_key('actions'):
actions = test_data['actions']
else:
actions = ['end_action','lchange','rchange','lturn','rturn']
# Prediction
rnn = []
if model_type == 'multipleRNNs':
if len(path_to_load_from) > 0:
rnn = loadMultipleRNNsCombined('{0}/checkpoint.{1}'.format(path_to_load_from,checkpoint))
else:
rnn = loadMultipleRNNsCombined('{2}/{0}/checkpoint.{1}'.format(index,checkpoint,path_to_checkpoints))
else:
if len(path_to_load_from) > 0:
rnn = load('{0}/checkpoint.{1}'.format(path_to_load_from,checkpoint))
else:
rnn = load('{2}/{0}/checkpoint.{1}'.format(index,checkpoint,path_to_checkpoints))
predictions = []
errors = 0
N = 0
P = []
Y = []
Time_before_maneuver = []
for xte,yte in zip(X_te,Y_te):
inputD = xte.shape[2]
road_feature_dimension = 4
prediction = []
if model_type == 'multipleRNNs':
prediction = rnn.predict_output([xte[:,:,(inputD-road_feature_dimension):],xte[:,:,:(inputD-road_feature_dimension)]],OutputActionThresh)
#[:,:,:(inputD-road_feature_dimension)]
else:
prediction = rnn.predict_output(xte,OutputActionThresh)
#print prediction.T
predictions.append(prediction)
t = np.nonzero(yte-prediction)
# Label 0 is the dummy label. Currently maneuvers are labeled [1..n]
prediction = prediction[:,0]
actual = yte[:,0]
prediction[prediction>0] -= 1
actual[actual>0] -= 1
p,anticipation_time = predictManeuver(prediction,actions)
y = actual[-1]
P.append(p)
Y.append(y)
Time_before_maneuver.append(anticipation_time)
result = {'actual':y,'prediction':p,'timeseries':list(prediction)}
#print result.values()
errors += len(t[0])
N += yte.shape[0]
P = np.array(P)
Y = np.array(Y)
Time_before_maneuver = np.array(Time_before_maneuver)
[conMat,p_mat,re_mat,time_mat] = confusionMat(P,Y,Time_before_maneuver)
return conMat,p_mat,re_mat,time_mat
def load_rnn(index,fold,checkpoint):
path_to_dataset = '/scr/ashesh/brain4cars/dataset/{0}'.format(fold)
path_to_checkpoints = '/scr/ashesh/brain4cars/checkpoints/{0}'.format(fold)
rnn = load('{2}/{0}/checkpoint.{1}'.format(index,checkpoint,path_to_checkpoints))
return rnn
layers = [TemporalInputFeatures(inputD),LSTM('tanh','sigmoid','orthogonal',4,512),softmax(num_classes)]
trY = T.lmatrix()
# Initializing network
rnn = RNN(layers,softmax_loss,trY,1e-3)
if not os.path.exists('{1}/{0}/'.format(index,path_to_checkpoints)):
os.mkdir('{1}/{0}/'.format(index,path_to_checkpoints))
# Fitting model
rnn.fitModel(X_tr,Y_tr,1,'{1}/{0}/'.format(index,path_to_checkpoints),epochs,batch_size,learning_rate_decay,decay_after)
else:
checkpoint = sys.argv[3]
# Prediction
rnn = load('{2}/{0}/checkpoint.{1}'.format(index,checkpoint,path_to_checkpoints))
if train_more:
rnn.fitModel(X_tr,Y_tr,1,'{1}/{0}/'.format(index,path_to_checkpoints),epochs,batch_size,learning_rate_decay,decay_after)
predictions = []
errors = 0
N = 0
for xte,yte in zip(X_te,Y_te):
prediction = rnn.predict_output(xte,OutputMaxProb)
predictions.append(prediction)
t = np.nonzero(yte-prediction)
print t
errors += len(t[0])
N += yte.shape[0]
cPickle.dump(predictions,open('{1}/prediction_{0}.pik'.format(index,path_to_dataset),'wb'))
