def run_svm_using_test_train_split(data, data_with_features, split, task):
"""
#
@param X: numpy
@param y: numpy
@return:
"""
# split training set and test set
(x_train, y_train), (x_test, y_test) = data.split_train_test(split,
stratify=True,
task=task)
# reset_train_test_split=True,
# splitted_edges_dir=
# split_by_node=True)
x_train_with_features, x_test_with_features = data_with_features.loc[
x_train], \
data_with_features.loc[
x_test]
# train model
clf = svm.SVC(gamma='scale',
decision_function_shape='ovr',
probability=True)
clf.fit(x_train_with_features, y_train)
## train set
### use decision_function
y_train_pred = clf.decision_function(x_train_with_features).argmax(1)
y_train_pred_proba = clf.decision_function(x_train_with_features)
## test set
### use decision function
y_test_pred = clf.decision_function(x_test_with_features).argmax(1)
y_test_pred_proba = clf.decision_function(x_test_with_features)
# report performance of model
print('=======training set=======')
train_report_np, train_columns, train_index = report_performance(
y_train,
y_train_pred,
y_train_pred_proba,
np.unique(y_train),
plot=True,
verbose=True,
return_value_for_cv=True)
print('=======test set=======')
test_report_np, test_columns, test_index = report_performance(
y_test,
y_test_pred,
y_test_pred_proba,
np.unique(y_test),
plot=True,
verbose=True,
return_value_for_cv=True)
#========= save to file=========
save2file(train_report_np, train_columns, train_index, test_report_np,
test_columns, test_index)
def run_rf_using_test_train_split(data, data_with_features, split, task):
"""
#
@param X: numpy
@param y: numpy
@return:
"""
# split training set and test set
(x_train, y_train), (x_test, y_test) = data.split_train_test(split,
stratify=True,
task=task)
# reset_train_test_split=True,
# split_by_node=True)
x_train_with_features, x_test_with_features = data_with_features.loc[
x_train], \
data_with_features.loc[
x_test]
# train model
clf = RandomForestClassifier(max_depth=2, random_state=0)
clf.fit(x_train_with_features, y_train)
## train set
### use decision_function
y_train_pred = clf.predict_proba(x_train_with_features).argmax(1)
y_train_pred_proba = clf.predict_proba(x_train_with_features)
## test set
### use decision function
y_test_pred = clf.predict_proba(x_test_with_features).argmax(1)
y_test_pred_proba = clf.predict_proba(x_test_with_features)
# report performance of model
print('=======training set=======')
train_report_np, train_columns, train_index = report_performance(
y_train,
y_train_pred,
y_train_pred_proba,
np.unique(y_train),
plot=True,
verbose=True,
return_value_for_cv=True)
print('=======test set=======')
test_report_np, test_columns, test_index = report_performance(
y_test,
y_test_pred,
y_test_pred_proba,
np.unique(y_test),
plot=True,
verbose=True,
return_value_for_cv=True)
#========= save to file=========
save2file(train_report_np, train_columns, train_index, test_report_np,
test_columns, test_index)
def run_neural_network_using_train_test_split(data,
x_with_features,
task,
splitted_edges_dir,
split=None,
split_by_node=None):
# =====================
# ==parameter setting
# =====================
# epochs = 1
# epochs = 10
# epochs = 150
# TODO make sure that x_with_features have row = edges and columns = features dimension
train_set_np, test_set_np = data.split_train_test(
split,
stratify=True,
task=task,
splitted_edges_dir=splitted_edges_dir,
split_by_node=split_by_node)
def get_train_test_set():
# train_set_np, test_set_np = data.train_set, data.test_set
# x_train,y_train, x_test,y_test have to be used as row index to get access to x_with_features
x_train_ind, y_train = train_set_np[:, :2], train_set_np[:, -1].astype(
float)
x_test_ind, y_test = test_set_np[:, :2], test_set_np[:,
-1].astype(float)
# from keras.utils import to_categorical
# y_train = to_categorical(y_train.astype(int))
# y_test = to_categorical(y_test.astype(int))
import numpy as np
# what should train and test of link prediction be?
x_with_features.index = x_with_features.index.map(str)
tmp = x_with_features.reindex(x_train_ind[:, 0]).dropna().to_numpy()
tmp_1 = x_with_features.reindex(x_train_ind[:, 1]).dropna().to_numpy()
x_train = np.concatenate([tmp, tmp_1], axis=1)
tmp = x_with_features.reindex(x_test_ind[:, 0]).dropna().to_numpy()
tmp_1 = x_with_features.reindex(x_test_ind[:, 1]).dropna().to_numpy()
x_test = np.concatenate([tmp, tmp_1], axis=1)
return (y_train, x_train, x_train_ind), (y_test, x_test, x_test_ind)
(y_train, x_train, x_train_ind), (y_test, x_test,
x_test_ind) = get_train_test_set()
# x_train = np.concatenate([x_with_features.loc[x_train_ind[:, 0]],
# x_with_features.loc[x_train_ind[:, 1]]], axis=1)
#
# x_test = np.concatenate([x_with_features.loc[x_test_ind[:, 0]],
# x_with_features.loc[x_test_ind[:, 1]]], axis=1)
assert x_train.shape[0] == x_train_ind.shape[0], ''
# TODO pass the correct value in
y_train_pred, y_train_pred_proba, y_test_pred, y_test_pred_proba = run_neural_network_for_each_fold(
x_train, x_test, y_train, y_test)
print(f"================training set==================")
report_final_train_performance_report_np, columns_of_performance_metric, indices_of_performance_metric = report_performance(
y_train,
y_train_pred,
y_train_pred_proba,
np.unique(y_train),
verbose=True,
plot=False,
return_value_for_cv=True)
print(f"================test set==================")
report_final_test_performance_report_np, columns_of_performance_metric, indices_of_performance_metric = report_performance(
y_test,
y_test_pred,
y_test_pred_proba,
np.unique(y_test),
verbose=True,
plot=False,
return_value_for_cv=True)
save2file(report_final_train_performance_report_np,
columns_of_performance_metric, indices_of_performance_metric,
report_final_test_performance_report_np,
columns_of_performance_metric, indices_of_performance_metric)
def run_lr_using_test_train_split_with_link_prediction(data, x_with_features,
task,
splitted_edges_dir,
split, split_by_node):
train_set_np, test_set_np = data.split_train_test(
split,
stratify=True,
task=task,
splitted_edges_dir=splitted_edges_dir,
split_by_node=split_by_node)
# train_set_np, test_set_np = data.train_set, data.test_set
# x_train,y_train, x_test,y_test have to be used as row index to get access to x_with_features
x_train_ind, y_train = train_set_np[:, :2], train_set_np[:,
-1].astype(float)
x_test_ind, y_test = test_set_np[:, :2], test_set_np[:, -1].astype(float)
# from keras.utils import to_categorical
# y_train = to_categorical(y_train.astype(int))
# y_test = to_categorical(y_test.astype(int))
import numpy as np
# what should train and test of link prediction be?
x_with_features.index = x_with_features.index.map(str)
tmp = x_with_features.reindex(x_train_ind[:, 0]).dropna().to_numpy()
tmp_1 = x_with_features.reindex(x_train_ind[:, 1]).dropna().to_numpy()
x_train = np.concatenate([tmp, tmp_1], axis=1)
tmp = x_with_features.reindex(x_test_ind[:, 0]).dropna().to_numpy()
tmp_1 = x_with_features.reindex(x_test_ind[:, 1]).dropna().to_numpy()
x_test = np.concatenate([tmp, tmp_1], axis=1)
# x_train = np.concatenate([x_with_features.loc[x_train_ind[:, 0]],
# x_with_features.loc[x_train_ind[:, 1]]], axis=1)
#
# x_test = np.concatenate([x_with_features.loc[x_test_ind[:, 0]],
# x_with_features.loc[x_test_ind[:, 1]]], axis=1)
assert x_train.shape[0] == x_train_ind.shape[0], ''
# train model
clf = LogisticRegression(random_state=0)
clf.fit(x_train, y_train)
## train set
### use decision_function
y_train_pred = clf.predict_proba(x_train).argmax(1)
y_train_pred_proba = clf.predict_proba(x_train)
## test set
### use decision function
y_test_pred = clf.predict_proba(x_test).argmax(1)
y_test_pred_proba = clf.predict_proba(x_test)
# report performance of model
print('=======training set=======')
train_report_np, train_columns, train_index = report_performance(
y_train,
y_train_pred,
y_train_pred_proba,
np.unique(y_train),
plot=True,
verbose=True,
return_value_for_cv=True)
print('=======test set=======')
test_report_np, test_columns, test_index = report_performance(
y_test,
y_test_pred,
y_test_pred_proba,
np.unique(y_test),
plot=True,
verbose=True,
return_value_for_cv=True)
#========= save to file=========
save2file(train_report_np, train_columns, train_index, test_report_np,
test_columns, test_index)