def run_xgboost_classifier(load=False, model_no=1):
# para load: whether or not to load pre-trained model
# para model_no: if load, which model to load
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
if load:
print("\nload pre-trained no.%d model" % model_no)
xgb_model = load_model(model_no)
else:
print("\ntraining a XGBoost classifier ...")
xgb_model = xgb.XGBClassifier(min_child_weight=MIN_CHILD_WEIGHT,
gamma=G,
subsample=SUBSAMPLE,
n_estimators=NO_TREES,
max_depth=MAX_DEPTH)
xgbt_name = "min_child_weight=%s,gamma=%s,subample=%s,n_estimators=%s,max_depth=%s" % (
str(MIN_CHILD_WEIGHT), str(G), str(SUBSAMPLE), str(NO_TREES),
str(MAX_DEPTH))
xgb_model.fit(X_train, y_train)
print("\ntraining DONE. \n\nsaving the XGBoost classifer ...")
save_model(xgb_model, xgbt_name)
print("\ntesting the XGBoost classifier ...")
y_pred = xgb_model.predict(X_test)
print(mean_squared_error(y_test, y_pred))
print(confusion_matrix(y_test, y_pred))
# show & save the top 50 important features
show_important_feature(xgb_model, data_RNASeq, img=False)
# draw the precision recall curve for the classifier
draw_precision_recall_curve(y_test, y_pred)
def run_random_forest(load=False, model_no=1):
# para load: whether or not to load pre-trained model
# para model_no: if load, which model to load
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
if load:
print("\nload pre-trained no.%d model" % model_no)
forest = load_model(model_no)
else:
print("\ntraining a Random Forest classifier ...")
forest = RandomForestClassifier(n_estimators=NO_TREES,
random_state=0,
max_features=MAX_FEATURES,
criterion=CRITERION,
n_jobs=NO_JOBS)
forest_name = "n_estimators=%s,max_features=%s,criterion=%s,n_jobs=%s" % (
NO_TREES, MAX_FEATURES, CRITERION, NO_JOBS)
forest.fit(X_train, y_train)
print("\ntraining DONE.\n\nsaving the RF classifier ...")
save_model(forest, forest_name)
print("\ntesting the Random Forest classifier ...\n")
y_pred = forest.predict(X_test)
print("Accuracy on training set: %.3f" % forest.score(X_train, y_train))
print("Accuracy on test set: %.3f" % forest.score(X_test, y_test))
# show & save the top 50 important features
show_important_feature(forest, data_RNASeq, img=False)
# draw the precision recall curve for the classifier
draw_precision_recall_curve(y_test, y_pred)
def run_gradient_boost(load=False, model_no=1):
# para load: whether or not to load pre-trained model
# para model_no: if load, which model to load
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
if load:
print("\nload pre-trained no.%d model" % model_no)
gbrt = load_model(model_no)
else:
print("\ntraining a Gradient Boosting Tree classifier ...")
gbrt = GradientBoostingClassifier(n_estimators=NO_TREES,
random_state=0,
max_features=MAX_FEATURES,
max_depth=MAX_DEPTH,
learning_rate=LEARNING_RATE)
gbrt_name = "n_estimators=%s,max_features=%s,max_depth=%s,learning_rate=%s" % (
str(NO_TREES), MAX_FEATURES, str(MAX_DEPTH), str(LEARNING_RATE))
gbrt.fit(X_train, y_train)
print("\ntraining DONE.\n\nsaving the GB classifier ...")
save_model(gbrt, gbrt_name)
print("\ntesting the Gradient Boosting Tree classifier ...\n")
y_pred = gbrt.predict(X_test)
print("Accuracy on training set: %.3f" % gbrt.score(X_train, y_train))
print("Accuracy on test set: %.3f" % gbrt.score(X_test, y_test))
# show & save the top 50 important features
show_important_feature(gbrt, data_RNASeq, img=False)
# draw the precision recall curve for the classifier
draw_precision_recall_curve(y_test, y_pred)
def tune_hyperparameters():
# load the data
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
parameters = {
"n_estimators": [200, 400, 500, 600, 800],
"max_features": ['log2', 'sqrt'],
"max_depth": [3, 5, 8],
"learning_rate": [0.05, 0.1, 0.2]
}
print(
"\nrunning the Grid Search for Gradient Boosting Tree classifier ...")
clf = GridSearchCV(GradientBoostingClassifier(),
parameters,
cv=2,
n_jobs=NO_JOBS,
verbose=10)
clf.fit(X_train, y_train)
print(clf.score(X_train, y_train))
print(clf.best_params_)
# save tune hyperparameters
with smart_open("./results/best_params_gradient_boost.txt",
'w',
encoding='utf-8') as f:
f.write(str(clf.best_params_) + str(clf.best_score_))
print("\nbest hyperparameters for GBRT has been written to file.")
def tune_hyperparameters():
# loda the data
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
params = {
"min_child_weight": [1, 5, 10],
"gamma": [0.5, 1, 2],
"subsample": [0.6, 0.8, 1.0],
"max_depth": [3, 5],
"n_estimators": [50, 200]
}
print("\nrunning the Grid Search for XGBoost classifier ...")
clf = GridSearchCV(xgb.XGBClassifier(),
params,
cv=2,
n_jobs=NO_JOBS,
verbose=10)
clf.fit(X_train, y_train)
print(clf.best_score_)
print(clf.best_estimator_)
# save tuned hyperparameters
with smart_open("./results/best_params_xgboost.txt", 'w',
encoding='utf-8') as f:
f.write(str(clf.best_params_) + str(clf.best_score_))
print("\nbest hyperparameters for XGBOOST has been written to file.")
def tune_hyperparameters():
# load the data
data_RNASeq_labels = load_data_RNASeq(proc=False,
label=False,
raw_count=True)
print(data_RNASeq_labels.iloc[1:3])
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
data_labels = data_RNASeq_labels['label']
data_RNASeq = data_RNASeq_labels.drop(columns=['label'])
# train/test split
print("\nsplitting the training/test dataset ...")
X_train, X_test, y_train, y_test = train_test_split(
data_RNASeq, data_labels)
parameters = {
"n_estimators": [100, 200, 300, 400, 500],
"max_features": [0.1, 0.2, 0.25, 0.3, 0.4, 0.5],
"criterion": ["entropy"]
}
print("\nrunning the Grid Search for Random Forest classifier ...")
clf = GridSearchCV(RandomForestClassifier(),
parameters,
cv=2,
n_jobs=NO_JOBS,
verbose=10)
clf.fit(X_train, y_train)
print(clf.score(X_train, y_train))
print(clf.best_params_)
# save tuned hyperparameters
with smart_open("./results/best_params_random_forest.txt",
'w',
encoding='utf-8') as f:
f.write(str(clf.best_params_) + str(clf.best_score_))
print("\nbest hyperparameters for RF has been written to file.")
def survival_analysis_with_all_RNASeq(model_type):
# para model_type:
data_RNASeq_labels = load_data_RNASeq()
data_RNASeq_labels = data_RNASeq_labels.drop(columns=['gene'])
feature_list = [] # list of gene signatures
# load most important features (index and name)
if model_type == 'rf':
for line in smart_open(IMPORTANT_FEATURE_RANDOM_FOREST,
'r',
encoding='utf-8'):
line = line.split()
feature_list.append((line[2], line[3]))
elif model_type == 'gbrt':
for line in smart_open(IMPORTANT_FEATURE_GRADIENT_BOOST,
'r',
encoding='utf-8'):
line = line.split()
feature_list.append((line[2], line[3]))
elif model_type == 'xgbt':
for line in smart_open(IMPORTANT_FEATURE_XGBOOST,
'r',
encoding='utf-8'):
line = line.split()
feature_list.append((line[2], line[3]))
else:
print(
"\nPlease indicate the type of model you have trained to produce important features"
)
log_p_values = []
for i in range(len(feature_list)):
log_p_values.append(
survival_analysis_with_one_RNASeq(model_type, data_RNASeq_labels,
feature_list, i))
print(log_p_values)
save_log_p_values(model_type, log_p_values)