def run(no_folds = 5, supervised = True, i = 0, l = 2, n = 128, expParameters = {}):
inst = KFold(n_splits = no_folds, shuffle=True, random_state=125)
exp = experiment.GGCNNExperiment('2018-08-28-SA1SA2', '2018-08-28-SA1SA2', SA1Experiment(neurons = n, blocks = l, **expParameters))
exp.num_iterations = 5000
exp.optimizer = 'adam'
exp.loss_type = 'linear'
exp.debug = True # Was True
exp.preprocess_data(dataset)
valid_idx = np.flatnonzero(dataset['level_0']['labels'] >= 0) # Missing data labelled with -1
if supervised:
train_idx, test_idx = list(inst.split( valid_idx ))[i]
else:
test_idx, train_idx = list(inst.split( valid_idx ))[i] # Reversed to get more samples in the test set than the training set
n_components = expParameters.get('linkage_adjustment_components', None)
exp.create_data(train_idx, test_idx, n_components = n_components)
exp.build_network()
results = exp.run()
# Node type of input nodes: 0 = training set; 1 = test set; -1 = neither
idx_split = np.empty((len(dataset['level_0']['labels']), 1))
idx_split.fill(-1)
idx_split[train_idx] = 0
idx_split[test_idx] = 1
return results, idx_split
def run(no_folds = 5, supervised = True, i = 0, l = 2, n = 128, expParameters = {}):
inst = KFold(n_splits = no_folds, shuffle=True, random_state=125)
exp = experiment.GGCNNExperiment('2018-08-28-SA1SA2', '2018-08-28-SA1SA2', SA1Experiment(neurons = n, blocks = l, **expParameters))
exp.num_iterations = 5000
exp.optimizer = 'adam'
exp.loss_type = 'linear'
exp.debug = True # Was True
exp.preprocess_data(dataset)
if supervised:
train_idx, test_idx = list(inst.split(np.arange(len(dataset[0]))))[i]
else:
test_idx, train_idx = list(inst.split(np.arange(len(dataset[0]))))[i] # Reversed to get more samples in the test set than the training set
exp.create_data(train_idx, test_idx)
exp.build_network()
results = exp.run()
return results
i = int(sys.argv[3])
except IndexError:
l = 2
n = 128
i = 0
saveName = 'Output/SemiSupervisedSydney-NoEdge-l={:d}-n={:d}-i={:d}.csv'.format(
l, n, i)
max_acc = []
iteration = []
layers = []
neurons = []
rep = []
exp = experiment.GGCNNExperiment('2018-06-06-SA1', '2018-06-06-sa1',
SA1Experiment(neurons=n, blocks=l))
# exp = experiment.SingleGraphCNNExperiment('2018-06-06-SA1', '2018-06-06-sa1', SA1Experiment(neurons = n, blocks = l))
exp.num_iterations = 1000
exp.optimizer = 'adam'
exp.debug = True # Was True
exp.preprocess_data(dataset)
train_idx, test_idx = list(inst.split(np.arange(len(dataset[-1]))))[i]
# test_idx, train_idx = list(inst.split(np.arange(len(dataset[-1]))))[i] # Reversed to get more samples in the test set than the training set
exp.create_data(train_idx, test_idx)
exp.build_network()
results = exp.run()
net.make_embedding_layer(self.neurons)
net.make_graphcnn_layer(1,
name='final',
with_bn=False,
with_act_func=False)
no_folds = 5 ##
inst = KFold(n_splits=no_folds, shuffle=True, random_state=125)
l = 2
n = 64
i = 2
exp = experiment.GGCNNExperiment('2018-08-28-SA1SA2', '2018-08-28-SA1SA2',
SA1Experiment(neurons=n, blocks=l))
exp.num_iterations = 2000
exp.optimizer = 'adam'
exp.loss_type = "linear"
exp.debug = True # Was True
exp.preprocess_data(dataset)
train_idx, test_idx = list(inst.split(np.arange(len(dataset[0]))))[i]
# print('Before: ', exp.train_idx.shape)
# exp.train_idx = np.append(exp.train_idx, np.arange( SA1DatasetSize , len(dataset[-1] )))
# exp.test_idx = np.append(exp.test_idx, np.arange( SA1DatasetSize , len(dataset[-1] )))
# print('After: ', exp.train_idx.shape)
# test_idx, train_idx = list(inst.split(np.arange(len(dataset[-1]))))[i] # Reversed to get more samples in the test set than the training set
