# preload some data just to get the dimensions
data[0].objects['train']['singletons'].load(memory=False)
dims = data[0].objects['train']['singletons'].data.shape
dims = (
None,
3,
)
'''
first build the classifier!
'''
# set up data
variables = ['tau32', 'tau21', 'msd']
classifier_train_gen = obj.generateSingletons(data,
variables,
partition='train',
batch=100)
classifier_validation_gen = obj.generateSingletons(data,
variables,
partition='validate',
batch=100)
classifier_test_gen = obj.generateSingletons(data,
variables,
partition='test',
batch=1000)
test_i, test_o, test_w = next(classifier_test_gen)
inputs = Input(shape=(dims[1], ), name='input')
dense = Dense(32,
activation='relu',
name='dense1',
kernel_initializer='lecun_uniform')(inputs)
dense = Dense(10, activation='tanh', kernel_initializer='lecun_uniform')(dense)
category_pred = Dense(config.n_truth,
activation='softmax',
kernel_initializer='lecun_uniform')(dense)
classifier = Model(inputs=inputs, outputs=category_pred)
classifier.compile(optimizer=Adam(lr=0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
classifier.summary()
classifier_train_gen = obj.generateSingletons(data,
partition='train',
batch=1000)
def save_model(name='classifier'):
classifier.save('%s.h5' % name)
def save_and_exit(signal=None, frame=None):
save_model()
flog.close()
exit(1)
# ctrl+C now triggers a graceful exit
signal.signal(signal.SIGINT, save_and_exit)
coll = obj.PFSVCollection()
coll.add_categories(['singletons'], fpath)
return coll
top = make_coll('/data/t3serv014/snarayan/deep/v_deep_2/PARTITION/Top_*_CATEGORY.npy')
hig = make_coll('/data/t3serv014/snarayan/deep/v_deep_2/PARTITION/Higgs_*_CATEGORY.npy')
qcd = make_coll('/data/t3serv014/snarayan/deep/v_deep_2/PARTITION/QCD_*_CATEGORY.npy')
data = [top, hig, qcd]
'''
first build the classifier!
'''
# set up data
classifier_train_gen = obj.generateSingletons(data, partition='train', batch=1000)
classifier_validation_gen = obj.generateSingletons(data, partition='validate', batch=10000)
classifier_test_gen = obj.generateSingletons(data, partition='validate', batch=10)
opts = {
'decorr_mass' : DECORRMASS,
'decorr_pt' : DECORRPT,
'decorr_rho' : DECORRRHO,
}
adv_train_gen = obj.generateSingletons(data, partition='train', batch=1000,**opts)
adv_validation_gen = obj.generateSingletons(data, partition='validate', batch=10000,**opts)
adv_test_gen = obj.generateSingletons(data, partition='validate', batch=10,**opts)
test_i, test_o, test_w = next(classifier_test_gen)
#print test_i
inputs = Input(shape=(len(obj.default_variables),), name='input')
dense = Dense(32, activation='tanh',name='dense1',kernel_initializer='lecun_uniform') (inputs)