def get_hp_space():
super_space = Model_lop.get_hp_space()
space = {
'n_hidden':
hp.choice('n_hidden', [
[
hopt_wrapper.qloguniform_int('n_hidden_1_' + str(i),
log(100), log(5000), 10)
for i in range(1)
],
[
hopt_wrapper.qloguniform_int('n_hidden_2_' + str(i),
log(100), log(5000), 10)
for i in range(2)
],
[
hopt_wrapper.qloguniform_int('n_hidden_3_' + str(i),
log(100), log(5000), 10)
for i in range(3)
],
]),
}
space.update(super_space)
return space
def get_hp_space():
super_space = Model_lop.get_hp_space()
space = {
'n_hidden': hopt_wrapper.qloguniform_int('n_hidden', log(100), log(5000), 10),
'n_factor': hopt_wrapper.qloguniform_int('n_factor', log(200), log(1000), 10),
}
space.update(super_space)
return space
def get_hp_space():
super_space = Model_lop.get_hp_space()
space = {
'num_filter_piano':
list_hopt_fixedSized([(20, 30, 1), (10, 20, 1)],
'num_filter_piano'),
'kernel_size_piano':
list_hopt_fixedSized([(12, 24, 1), (12, 24, 1)],
"kernel_size_piano"),
'num_filter_orch':
list_hopt_fixedSized([(30, 50, 1), (10, 20, 1)],
'num_filter_orch'),
'kernel_size_orch':
list_hopt_fixedSized([(12, 24, 1), (12, 24, 1)],
"kernel_size_orch"),
'embeddings_size':
qloguniform_int("embeddings_size", log(500), log(2000), 10),
'mlp_pred':
list_log_hopt(500, 2000, 10, 1, 3, "mlp_pred"),
'gru_orch':
list_log_hopt(500, 2000, 10, 0, 2, "gru_orch"),
}
space.update(super_space)
return space
def get_hp_space():
super_space = MLFPP.get_hp_space()
space = {
'hs_piano': list_log_hopt(500, 2000, 10, 0, 2, 'hs_piano'),
'hs_orch': list_log_hopt(500, 2000, 10, 0, 2, 'hs_orch'),
'embeddings_size': qloguniform_int('hs_orch', log(500), log(1000), 10),
}
space.update(super_space)
return space
def get_hp_space():
super_space = MLFPP.get_hp_space()
space = {
'recurrent_orch':
hp.choice('n_hidden', [
[
hopt_wrapper.qloguniform_int('n_hidden_1_' + str(i),
log(500), log(3000), 10)
for i in range(1)
],
[
hopt_wrapper.qloguniform_int('n_hidden_2_' + str(i),
log(500), log(3000), 10)
for i in range(2)
],
[
hopt_wrapper.qloguniform_int('n_hidden_3_' + str(i),
log(500), log(3000), 10)
for i in range(3)
],
]),
'recurrent_piano':
hp.choice('n_hidden', [
[
hopt_wrapper.qloguniform_int('n_hidden_1_' + str(i),
log(500), log(3000), 10)
for i in range(1)
],
[
hopt_wrapper.qloguniform_int('n_hidden_2_' + str(i),
log(500), log(3000), 10)
for i in range(2)
],
]),
'mlp_piano':
hp.choice('n_hidden', [
[
hopt_wrapper.qloguniform_int('n_hidden_1_' + str(i),
log(500), log(3000), 10)
for i in range(1)
],
[
hopt_wrapper.qloguniform_int('n_hidden_2_' + str(i),
log(500), log(3000), 10)
for i in range(2)
],
])
}
space.update(super_space)
return space
def get_hp_space():
space_training = {
'temporal_order':
hopt_wrapper.qloguniform_int('temporal_order', log(3), log(20), 1)
}
space_regularization = {
'dropout_probability':
hp.choice('dropout',
[0.0, hp.normal('dropout_probability', 0.5, 0.1)]),
'weight_decay_coeff':
hp.choice('weight_decay_coeff',
[0.0, hp.uniform('a', 1e-4, 1e-4)]),
}
space_training.update(space_regularization)
return space_training
def get_hp_space():
space_training = {
# 'temporal_order': hopt_wrapper.qloguniform_int('temporal_order', log(2), log(7), 1),
'temporal_order':
hopt_wrapper.qloguniform_int('temporal_order', log(5), log(5), 1),
'tn_weight':
1 / 10,
'sparsity_coeff':
0,
}
space_regularization = {
'dropout_probability': hp.choice('dropout', [0.0]),
'weight_decay_coeff': hp.choice('weight_decay_coeff', [0.0])
}
space_training.update(space_regularization)
return space_training
def get_hp_space():
super_space = Model_lop.get_hp_space()
space = {
'n_hidden_embedding': hp.choice('n_hidden_embedding', [
[hopt_wrapper.qloguniform_int('n_hidden_embedding_'+str(i), log(1500), log(3000), 10) for i in range(1)],
[hopt_wrapper.qloguniform_int('n_hidden_embedding_'+str(i), log(1500), log(3000), 10) for i in range(2)],
[hopt_wrapper.qloguniform_int('n_hidden_embedding_'+str(i), log(1500), log(3000), 10) for i in range(3)],
]),
'n_hidden_NADE': hp.choice('n_hidden_NADE', [
[hopt_wrapper.qloguniform_int('n_hidden_NADE_'+str(i), log(1500), log(3000), 10) for i in range(1)],
[hopt_wrapper.qloguniform_int('n_hidden_NADE_'+str(i), log(1500), log(3000), 10) for i in range(2)],
[hopt_wrapper.qloguniform_int('n_hidden_NADE_'+str(i), log(1500), log(3000), 10) for i in range(3)],
]),
'num_ordering': quniform_int('num_ordering', 5, 10, 1)
}
space.update(super_space)
return space
def list_log_hopt(min_unit, max_unit, step, min_num_layer, max_num_layer,
name):
return hp.choice(name, [
[qloguniform_int(name+'_'+str(i), log(min_unit), log(max_unit), step) for i in range(num_layer)] \
for num_layer in range(min_num_layer, max_num_layer)
])
