def save_adult_party_data(nb_dp_per_party, should_stratify, party_folder):
"""
Saves Adult party data
:param nb_dp_per_party: the number of data points each party should have
:type nb_dp_per_party: `list[int]`
:param should_stratify: True if data should be assigned proportional to source class distributions
:type should_stratify: `bool`
:param party_folder: folder to save party data
:type party_folder: `str`
"""
dataset_path = os.path.join("examples", "datasets")
if not os.path.exists(dataset_path):
os.makedirs(dataset_path)
x_train = load_adult(download_dir=dataset_path)
num_train = len(x_train.index)
y_train = x_train['class'].values.tolist()
labels, counts = np.unique(y_train, return_counts=True)
if should_stratify:
strat_col = y_train
groups, counts = np.unique(strat_col, return_counts=True)
# to use custom proportions, replace probs with a dictionary where key:value pairs are label:proportion
probs = {
group: counts[np.where(groups == group)[0][0]] / float(num_train)
for group in groups
}
p_list = np.array([probs[strat_col[idx]] for idx in range(num_train)])
p_list /= np.sum(p_list)
else:
probs = {label: 1.0 / num_train for label in labels}
p_list = np.array([probs[y_train[idx]] for idx in range(num_train)])
p_list /= np.sum(p_list)
for i, dp in enumerate(nb_dp_per_party):
# Create variable for indices
indices = np.random.choice(num_train, dp, p=p_list)
indices = indices.tolist()
# Use indices for data/classification subset
x_train_pi = x_train.iloc[indices]
name_file = 'data_party' + str(i) + '.csv'
name_file = os.path.join(party_folder, name_file)
with open(name_file, 'w') as writeFile:
writer = csv.writer(writeFile)
writer.writerows(x_train_pi)
x_train_pi.to_csv(path_or_buf=name_file, index=None)
print('Finished! :) Data saved in', party_folder)
def get_model_config(folder_configs, dataset, is_agg=False, party_id=0):
if dataset == 'adult':
loaded_data = load_adult()
# preproces the dataset first before generating the model spec
dh = AdultDTDataHandler()
loaded_data = dh.preprocess(loaded_data)
elif dataset == 'nursery':
loaded_data = load_nursery()
spec = dict()
spec['list_of_features'] = list(range(loaded_data.shape[1] - 1))
feature_values = list()
for feature in range(loaded_data.shape[1]):
if loaded_data.columns[feature] != 'class':
new_feature = loaded_data[loaded_data.columns[feature]
].cat.categories
feature_values.append(new_feature.tolist())
spec['feature_values'] = feature_values
list_of_labels = loaded_data['class'].cat.categories
spec['list_of_labels'] = list_of_labels.tolist()
f_spec = os.path.join(folder_configs, 'dt_model_spec.json')
with open(f_spec, 'w') as f:
json.dump(spec, f)
spec = {
'model_name': 'decision-tree',
'model_definition': os.path.join(folder_configs, 'dt_model_spec.json')
}
model = {
'name': 'DTFLModel',
'path': 'ibmfl.model.dt_fl_model',
'spec': spec
}
return model