def do_ssl_for_one_column(data, label_index, feat_indices, one_hot_cols,
min_max_cols, unlabel_num):
"""
Does semi-supervised learning experiments with the selected column as labels.
:param data: original data
:param label_index: label index
:param feat_indices: feature indices
:param one_hot_cols: columns for one-hot encoding
:param min_max_cols: columns for min-max normalization
:param unlabel_num: number of unlabeled data points
"""
data_X, data_y = split_feat_and_label(data=data,
label_index=label_index,
feat_indices=feat_indices)
data_X = one_hot_preprocess_sl(data_X=data_X,
one_hot_cols=one_hot_cols,
min_max_cols=min_max_cols)
train_X, train_y, test_X, test_y = get_tr_te_set(data_X=data_X,
data_y=data_y,
test_size=895)
"""
train_X: (4000, 32), train_y: (4000,)
test_X: (895, 32), test_y: (895,)
"""
do_ssl_experiments(train_X=train_X,
train_y=train_y,
unlabel_num=unlabel_num,
test_X=test_X,
test_y=test_y,
label_name=str(label_index))
def do_sl_for_one_column(data, label_index, feat_indices, one_hot_cols,
min_max_cols):
"""
Does supervised learning experiments with the selected column as labels.
:param data: original data.
:param label_index: label index.
:param feat_indices: feature indices.
:param one_hot_cols: columns for one-hot encoding.
:param min_max_cols: columns for min-max normalization.
"""
data_X, data_y = split_feat_and_label(data=data,
label_index=label_index,
feat_indices=feat_indices)
data_X = one_hot_preprocess_sl(data_X=data_X,
one_hot_cols=one_hot_cols,
min_max_cols=min_max_cols)
train_X, train_y, test_X, test_y = get_tr_te_set(data_X=data_X,
data_y=data_y,
test_size=895)
"""
train_X: (4000, 32), train_y: (4000,)
test_X: (895, 32), test_y: (895,)
"""
# Normal experiments
do_sl_experiments(train_X=train_X,
train_y=train_y,
test_X=test_X,
test_y=test_y,
label_name=str(label_index))
# Experiments for deleting the only numerical feature
do_sl_experiments(train_X=train_X[:, 0:31],
train_y=train_y,
test_X=test_X[:, 0:31],
test_y=test_y,
label_name=str(label_index) + '_nm')
def do_tl_for_one_column(data, label_index, feat_indices, split_col,
one_hot_cols, min_max_cols, test_size):
"""
Does transfer learning experiments with the selected column as labels.
:param data: original data
:param label_index: label index
:param feat_indices: feature indices
:param split_col: column to be split for transfer learning
:param one_hot_cols: columns for one-hot encoding
:param min_max_cols: columns for min-max normalization.
:param test_size: size of test data for target
"""
data_X, data_y = split_feat_and_label(data=data,
label_index=label_index,
feat_indices=feat_indices)
prepro_source_X, source_data_y, prepro_target_X, target_data_y = \
one_hot_preprocess_tl(data_X=data_X, data_y=data_y, split_col=split_col,
one_hot_cols=one_hot_cols, min_max_cols=min_max_cols)
source_X, source_y, target_tr_X, target_tr_y, target_te_X, target_te_y = \
get_so_ta_set(prepro_source_X=prepro_source_X, source_data_y=source_data_y,
prepro_target_X=prepro_target_X, target_data_y=target_data_y,
test_size=test_size)
do_tl_experiments(source_X=source_X,
source_y=source_y,
target_tr_X=target_tr_X,
target_tr_y=target_tr_y,
target_te_X=target_te_X,
target_te_y=target_te_y,
label_name=str(label_index),
split_name=str(split_col))
def get_sl_for_one_column(data, label_index, feat_indices, one_hot_cols,
min_max_cols):
"""
Gets supervised learning results with the selected column as labels.
:param data: original data.
:param label_index: label index.
:param feat_indices: feature indices.
:param one_hot_cols: columns for one-hot encoding.
:param min_max_cols: columns for min-max normalization.
"""
data_X, data_y = split_feat_and_label(data=data,
label_index=label_index,
feat_indices=feat_indices)
data_X = one_hot_preprocess_sl(data_X=data_X,
one_hot_cols=one_hot_cols,
min_max_cols=min_max_cols)
train_X, train_y, test_X, test_y = get_tr_te_set(data_X=data_X,
data_y=data_y,
test_size=895)
"""
train_X: (4000, 32), train_y: (4000,)
test_X: (895, 32), test_y: (895,)
"""
# Normal experiments
lp_filename = "saved_models/sl/linear perceptron " + str(
label_index) + ".pkl"
gnp_filename = "saved_models/sl/Gaussian Naive Bayes " + str(
label_index) + ".pkl"
svc_filename = "saved_models/sl/support vector machine " + str(
label_index) + ".pkl"
lgr_filename = "saved_models/sl/logistic regression " + str(
label_index) + ".pkl"
rf_filename = "saved_models/sl/random forest " + str(label_index) + ".pkl"
adb_filename = "saved_models/sl/AdaBoost " + str(label_index) + ".pkl"
get_sl_results(train_X=train_X,
train_y=train_y,
test_X=test_X,
test_y=test_y,
label_name=str(label_index),
lp_filename=lp_filename,
gnp_filename=gnp_filename,
svc_filename=svc_filename,
lgr_filename=lgr_filename,
rf_filename=rf_filename,
adb_filename=adb_filename)
# Experiments for deleting the only numerical feature
lp_filename_nm = "saved_models/sl/linear perceptron " + str(
label_index) + "_nm.pkl"
gnp_filename_nm = "saved_models/sl/Gaussian Naive Bayes " + str(
label_index) + "_nm.pkl"
svc_filename_nm = "saved_models/sl/support vector machine " + str(
label_index) + "_nm.pkl"
lgr_filename_nm = "saved_models/sl/logistic regression " + str(
label_index) + "_nm.pkl"
rf_filename_nm = "saved_models/sl/random forest " + str(
label_index) + "_nm.pkl"
adb_filename_nm = "saved_models/sl/AdaBoost " + str(
label_index) + "_nm.pkl"
get_sl_results(train_X=train_X[:, 0:31],
train_y=train_y,
test_X=test_X[:, 0:31],
test_y=test_y,
label_name=str(label_index) + '_nm',
lp_filename=lp_filename_nm,
gnp_filename=gnp_filename_nm,
svc_filename=svc_filename_nm,
lgr_filename=lgr_filename_nm,
rf_filename=rf_filename_nm,
adb_filename=adb_filename_nm)