param_dict=param_dict,
n_permutations=24,
kwarg_key_list=[
'conv_filters1', 'kernel_sizes', 'maps_per_kernel', 'dropout1',
'dropout2', 'pool_size', 'conv_filters2', 'dense_units', 'lr'
])
elif ACTION == 'train_final_model':
fncs.train_final_model()
elif ACTION == 'cross_validation':
fncs.run_cross_validation(nn_architechture='CNNPAR',
in_kwargs={
'conv_filters1': 75,
'conv_filters2': 300,
'dense_units': 120,
'dropout1': 0.0,
'dropout2': 0.0,
'kernel_sizes': [6, 8, 10],
'lr': 0.001,
'maps_per_kernel': 2,
'pool_size': 3
})
elif ACTION == 'test_final_model':
importlib.reload(fncs)
weights_path = (
"code_modules/nn_training/BAC_UNI_len2006/final_elmo_CNNPAR"
"_BAC_UNI_len2006/final_elmo_CNNPAR_BAC_UNI_len2006"
"_1565079880")
fncs.train_final_model(just_test_model=True, weights_path=weights_path)
else:
import importlib import code_modules.nn_training.functions as fncs importlib.reload(fncs) fncs.run_cross_validation(nn_architechture='CNN')
import importlib
import code_modules.nn_training.functions as fncs
importlib.reload(fncs)
DO_GRIDSEARCH = False
if DO_GRIDSEARCH:
param_dict = {
'l1_units': [32, 64, 320],
'l2_units': [32, 64, 320],
'dropout1': [0.0, 0.2, 0.5],
'dropout2': [0.0, 0.2, 0.5],
'regularizer_vals': [0, 0.01, 0.001],
'epochs': [20, 50, 100],
'batch_size': [32, 32 * 4, 32 * 8]
}
results = fncs.hyperparameter_gridsearch(param_dict=param_dict)
else:
fncs.run_cross_validation(nn_architechture='DNN',
in_kwargs=dict(dropout1=0.5,
dropout2=0.5,
l1_units=32,
l2_units=64,
regularizer_vals=0),
epochs=50,
batch_size=256)
import code_modules.nn_training.functions as fncs
fncs.run_cross_validation(nn_architechture='BIDGRU', do_w2v=True,
dataset_to_use='GLYLIP')
'dropout1': [0.2],
'dropout2': [0.0],
'epochs': [150],
'batch_size': [256]
}
results = fncs.hyperparameter_gridsearch(nn_architechture='RNNCNN',
param_dict=param_dict,
n_permutations=6,
kwarg_key_list=[
'lr', 'conv_filters',
'kernel_size', 'pool_size',
'lstm_units', 'dropout1',
'dropout2'
])
else:
fncs.run_cross_validation(nn_architechture='RNNCNN',
fold_i_to_skip=tuple(),
**{
'batch_size': 1024,
'epochs': 75,
'in_kwargs': {
'conv_filters': 640,
'dropout1': 0.2,
'dropout2': 0.0,
'kernel_size': 26,
'lr': 0.0001,
'pool_size': 26
}
})
import code_modules.nn_training.functions as fncs fncs.run_cross_validation(nn_architechture='BIDGRU')