def predict(flags_obj):
"""Run Wide-Deep training and eval loop.
Args:
flags_obj: An object containing parsed flag values.
"""
dirProject = '/home/thomas/fusessh/scicore/projects/patrec'
# dirProject = "Z:\\projects\\PATREC"
dirResultsBase = os.path.join(dirProject, 'results/')
dirData = os.path.join(dirProject, 'data')
dict_options_dataset_training = {
'dir_data': dirData,
'data_prefix': 'patrec',
'dataset': '20122015',
'grouping': 'verylightgrouping',
'encoding': 'embedding',
'newfeatures': None,
'featurereduction': None,
'filtering': 'EntlassBereich_Gyn',
'balanced': False,
'resample': False
}
dataset_options_training = DatasetOptions(dict_options_dataset_training)
dict_options_dataset_testing = {
'dir_data': dirData,
'data_prefix': 'patrec',
'dataset': '20162017',
'grouping': 'verylightgrouping',
'encoding': 'embedding',
'newfeatures': None,
'featurereduction': None,
'filtering': 'EntlassBereich_Gyn',
'balanced': False,
'resample': False
}
dataset_options_testing = DatasetOptions(dict_options_dataset_testing)
if dict_options_dataset_testing['data_prefix'] == 'nz':
feature_columns = FeatureColumnsNZ(
dataset_options=dataset_options_testing)
# feature_columns_nz_fusion = FeatureColumnsNZFusion(dataset_options=dataset_options_testing);
# feature_columns = feature_columns_nz_fusion;
elif dict_options_dataset_testing['data_prefix'] == 'patrec':
feature_columns = FeatureColumnsPatrec(
dataset_options=dataset_options_testing)
# feature_columns_patrec_fusion = FeatureColumnsPatrecFusion(dataset_options=dataset_options_testing);
# feature_columns = feature_columns_patrec_fusion;
else:
print('unknown data prefix..exit')
sys.exit()
dict_dataset_options = {
'train': dataset_options_training,
'eval': None,
'test': dataset_options_testing
}
nn = NeuralNetModel('test', dict_dataset_options, feature_columns,
flags_obj)
model_flags = nn.getFlags()
if model_flags.model_dir.endswith('/'):
trained_model = model_flags.model_dir.split('/')[-2]
else:
trained_model = model_flags.model_dir.split('/')[-1]
if trained_model.startswith('warmstart'):
pretrained = 'pretrained'
else:
pretrained = None
print('warmstart: ' + str(trained_model.startswith('warmstart')))
print('hidden units: ' + str(model_flags.hidden_units))
dict_options_nn = {
'hidden_units': model_flags.hidden_units,
'learningrate': model_flags.learningrate,
'dropout': model_flags.dropout,
'batchnorm': model_flags.batchnorm,
'batch_size': model_flags.batch_size,
'training_epochs': model_flags.train_epochs,
'pretrained': pretrained,
}
options_nn = OptionsNN(model_flags.model_dir,
dataset_options_training,
options_clf=dict_options_nn)
classifier_nn = ClassifierNN(options_nn)
results_all_runs_test = Results(dirResultsBase, dataset_options_training,
options_nn, 'test',
dataset_options_testing)
num_runs = 10
test_auc = []
test_avgprecision = []
for k in range(0, num_runs):
results = nn.predict()
filename_data_testing = nn.getFilenameDatasetBalanced()
df_testing_balanced = pd.read_csv(filename_data_testing)
predictions = [p['probabilities'] for p in results]
predictions = np.array(predictions)
print('get labels...: ' + str(filename_data_testing))
labels = df_testing_balanced[
dataset_options_testing.getEarlyReadmissionFlagname()].values
res = classifier_nn.setResults(predictions, labels)
results_all_runs_test.addResultsSingleRun(res)
auc = res.getAUC()
avgprecision = res.getAvgPrecision()
print('')
print('AUC: ' + str(auc))
print('avg precision: ' + str(avgprecision))
print('')
test_auc.append(auc)
test_avgprecision.append(avgprecision)
print('')
print('mean test auc: ' + str(np.mean(np.array(test_auc))))
print('mean test avg precision: ' +
str(np.mean(np.array(test_avgprecision))))
print('')
results_all_runs_test.writeResultsToFileDataset()
years = [2012, 2013, 2014, 2015]
for year in years:
dict_options_dataset_training = {
'dir_data': dirData,
'data_prefix': 'nz',
'dataset': str(year),
'newfeatures': {
'names': constantsNZ.NEW_FEATURES
},
'featurereduction': None
}
options_training = DatasetOptions(dict_options_dataset_training)
dataset_training = Dataset(dataset_options=options_training)
early_readmission_flagname = options_training.getEarlyReadmissionFlagname(
)
print('dataset filename: ' + str(dataset_training.getFilename()))
results_all_runs_train = Results(dirResultsBase, options_training,
options_sgd, 'train')
results_all_runs_eval = Results(dirResultsBase, options_training,
options_sgd, 'eval')
df_balanced_test = dataset_testing.getBalancedSubSet()
num_runs = 1
eval_aucs = []
for run in range(0, num_runs):
print('')
[df_balanced_train, df_balanced_eval
