class TestNode(unittest.TestCase):
def setUp(self):
X, y = make_classification(n_features=5, random_state=1)
self.classifier = XGBClassifier(n_estimators=3)
self.classifier.fit(X, y)
self.predictions = self.classifier.predict_proba(X)
self.model_dump = [
tree.split('\n') for tree in self.classifier.booster().get_dump()
]
def test_parse_root_node(self):
node = bdt2cpp.Node(self.model_dump[0][0])
self.assertEqual(node.cut_value, -0.464102)
self.assertFalse(node.weight)
self.assertEqual(node.root, node)
self.assertIsNone(node.parent)
return node
def test_parse_left_node(self):
node = self.test_parse_root_node()
root = node
node.left = bdt2cpp.Node(self.model_dump[0][1], parent=node)
node = node.left
self.assertEqual(node.parent, root)
self.assertFalse(node.cut_value)
self.assertEqual(node.weight, -0.184906)
return node
def get_xgb_feature_importance_plot(best_param_, experiment_,
png_folder,
png_fname,
score_threshold=0.8):
# 1.
train_X, train_y = experiment_.get_train_data()
clf = XGBClassifier()
try:
del best_param_['model_type']
except:
pass
clf.set_params(**best_param_)
clf.fit(train_X, train_y)
index2feature = clf.booster().get_fscore()
fis = pd.DataFrame({'name':index2feature.keys(),
'score':index2feature.values()})
fis = fis.sort('score', ascending=False)
if len(fis.index) > 20:
score_threshold = fis['score'][fis['score'] > 0.0].quantile(score_threshold)
#where_str = 'score > %f & score > %f' % (score_threshold, 0.0)
where_str = 'score >= %f' % (score_threshold)
fis = fis.query(where_str)
# 2. plot
#gs = GridSpec(2,2)
#ax1 = plt.subplot(gs[:,0])
#ax2 = plt.subplot(gs[0,1])
#ax3 = plt.subplot(gs[1,1])
# 3.1 feature importance
sns.barplot(x = 'score', y = 'name',
data = fis,
#ax=ax1,
color="blue")
#plt.title("Feature_Importance", fontsize=10)
plt.ylabel("Feature", fontsize=10)
plt.xlabel("Feature_Importance : f-Score", fontsize=10)
"""
# 3.2 PDF
confidence_score = clf.oob_decision_function_[:,1]
sns.distplot(confidence_score, kde=False, rug=False, ax=ax2)
ax2.set_title("PDF")
# 3.3 CDF
num_bins = min(best_param_.get('n_estimators',1), 100)
counts, bin_edges = np.histogram(confidence_score, bins=num_bins, normed=True)
cdf = np.cumsum(counts)
ax3.plot(bin_edges[1:], cdf / cdf.max())
ax3.set_title("CDF")
ax3.set_xlabel("Oob_Decision_Function:Confidence_Score", fontsize=10)
"""
png_fname = os.path.join(Config.get_string('data.path'), 'graph', png_fname)
plt.tight_layout()
plt.savefig(png_fname)#, bbox_inches='tight', pad_inches=1)
plt.close()
return True
def xgb_train_offline():
print('data process...')
data = get_data()
data = fea_select(data)
id_fea = ['user_id', 'item_id', 'shop_id', 'context_id', 'context_page_id']
data.drop(id_fea, axis=1, inplace=True) # 暂时不先利用这些id特征
train, val, test = gen_train_val_test(data, True)
y_train = train['is_trade']
X_train = train.drop(['is_trade'], axis=1)
y_val = val['is_trade']
X_val = val.drop(['is_trade'], axis=1)
print('start training...')
xgb = XGBClassifier(objective='binary:logistic',
learning_rate=0.01,
n_estimators=10,
max_depth=5,
subsample=0.7,
colsample_bytree=0.7,
reg_lambda=0.005,
nthread=4,
seed=128,
silent=10)
xgb.fit(X_train,
y_train,
eval_set=[(X_train, y_train), (X_val, y_val)],
eval_metric='logloss',
early_stopping_rounds=50)
y_prob = xgb.predict_proba(X_val, ntree_limit=xgb.best_ntree_limit)[:, 1]
print('result log_loss = {0}'.format(log_loss(y_val, y_prob)))
fea_score = pd.DataFrame()
fea_score['feature'] = X_train.columns.tolist()
fea_score['score'] = list(xgb.booster().get_fsocre())
fea_score = fea_score.sort_values(by='score',
ascending=False).reset_index(drop=True)
fea_score.to_csv(file_path + 'fea_score.csv', index=None)
fea_map = pd.Series(data=fea_score['score'].values,
index=fea_score['feature'])
fea_map.plot(kind='bar', title='Feature Importances')
plt.ylabel('Feature Importance Score')
plt.show()
print('end...')
def XGBooster(Xtrain, Ytrain, Xtest, Ytest):
# Train
XG = XGBClassifier()
XG.fit(Xtrain, Ytrain)
#Feature Importancess
XGFeature_importances = pd.DataFrame(XG.booster().get_fscore(),
index=Xtrain.columns,
columns=['importance']).sort_values(
'importance', ascending=False)
# Test
yPred = XG.predict(Xtest)
XGPredictions = [round(value) for value in yPred]
XGAccuracy = accuracy_score(Ytest, XGPredictions)
return XGFeature_importances, XG, yPred, XGPredictions, XGAccuracy, print(
"Accuracy: %.2f%%" % (XGAccuracy * 100.0))
# train
clf.fit(X, y, sample_weight=w)
#save results
if options.optimize:
with open('%s/best_params.json' % options.out_dir, 'w+') as fout:
fout.write(json.dumps(clf.best_params_))
pd.DataFrame(clf.cv_results_).to_hdf('%s/cv_results.hd5' % options.out_dir,
key='cv_results')
if options.refit:
clf = clf.best_estimator_
else:
with open('%s/best_params.json' % options.out_dir, 'w+') as fout:
fout.write(json.dumps(options.clf_params))
if not options.optimize or optimize.optimize and options.refit:
if options.save_pickle:
with gopen('%s/model.pkl.gz' % options.out_dir, 'w+') as fout:
pickle.dump(clf, fout)
fout.close()
try:
model = clf.get_booster()
except:
model = clf.booster()
model.save_model('%s/model.xgb' % options.out_dir)
##
##
## # train it
## clf.fit(X_train,y_train,w_train)
df_all = pd.concat([df_all, bow], axis=1)
df_all['num_zero'] = num_zero
df_all = pipeline.fit(df_all).transform(df_all)
X_train = df_all.iloc[:df_train.shape[0], :]
X_test = df_all.iloc[df_train.shape[0]:, :]
y_train = df_target
ID_test = df_id
# best params so far using column/row subsampling, even longer training
learning_rate = 0.01
n_estimators = 800
max_depth = 6
subsample = 0.9
colsample_bytree = 0.85
min_child_weight = 1 # default
xgb = XGBClassifier(seed=0, learning_rate=learning_rate, n_estimators=n_estimators,
min_child_weight=min_child_weight, max_depth=max_depth,
colsample_bytree=colsample_bytree, subsample=subsample)
xgb = xgb.fit(X_train, y_train, eval_set=[(X_train, y_train)], eval_metric='auc')
importances = xgb.booster().get_fscore()
df_importance = pd.DataFrame(zip(importances.keys(), importances.values()), columns=['feature', 'importance'])
print df_importance.sort_values('importance', ascending=False).reset_index(drop=True)
y_pred = xgb.predict_proba(X_test)
submission = pd.DataFrame({'ID': ID_test, 'TARGET': y_pred[:, 1]})
submission.to_csv(filename, index=False)
print 'Wrote %s' % filename
def get_xgb_feature_importance_plot(best_param_,
experiment_,
png_folder,
png_fname,
score_threshold=0.8):
# 1.
train_X, train_y = experiment_.get_train_data()
clf = XGBClassifier()
try:
del best_param_['model_type']
except:
pass
clf.set_params(**best_param_)
clf.fit(train_X, train_y)
index2feature = clf.booster().get_fscore()
fis = pd.DataFrame({
'name': index2feature.keys(),
'score': index2feature.values()
})
fis = fis.sort('score', ascending=False)
if len(fis.index) > 20:
score_threshold = fis['score'][fis['score'] > 0.0].quantile(
score_threshold)
#where_str = 'score > %f & score > %f' % (score_threshold, 0.0)
where_str = 'score >= %f' % (score_threshold)
fis = fis.query(where_str)
# 2. plot
#gs = GridSpec(2,2)
#ax1 = plt.subplot(gs[:,0])
#ax2 = plt.subplot(gs[0,1])
#ax3 = plt.subplot(gs[1,1])
# 3.1 feature importance
sns.barplot(
x='score',
y='name',
data=fis,
#ax=ax1,
color="blue")
#plt.title("Feature_Importance", fontsize=10)
plt.ylabel("Feature", fontsize=10)
plt.xlabel("Feature_Importance : f-Score", fontsize=10)
"""
# 3.2 PDF
confidence_score = clf.oob_decision_function_[:,1]
sns.distplot(confidence_score, kde=False, rug=False, ax=ax2)
ax2.set_title("PDF")
# 3.3 CDF
num_bins = min(best_param_.get('n_estimators',1), 100)
counts, bin_edges = np.histogram(confidence_score, bins=num_bins, normed=True)
cdf = np.cumsum(counts)
ax3.plot(bin_edges[1:], cdf / cdf.max())
ax3.set_title("CDF")
ax3.set_xlabel("Oob_Decision_Function:Confidence_Score", fontsize=10)
"""
png_fname = os.path.join(Config.get_string('data.path'), 'graph',
png_fname)
plt.tight_layout()
plt.savefig(png_fname) #, bbox_inches='tight', pad_inches=1)
plt.close()
return True
colsample_bytree=0.8, # 0.9
scale_pos_weight=14, # 10
objective="reg:logistic",
nthread=-1,
seed=random_seed)
# analyse_n_estimators(model, train_x, train_y, est_list=[8000, 9000, 10000, 11000, 12000])
# params = {"max_depth": [3, 4, 5],
# "min_child_weight": [1, 10, 100],}
# params = {"min_child_weight": [5, 7, 9, 10, 11, 13, 15]}
# params = {"subsample": [0.7, 0.8, 0.9, 1],
# "colsample_bytree": [0.7, 0.8, 0.9, 1],}
# params = {"reg_lambda": [0.1, 1, 10, 100]}
# params = {"scale_pos_weight": [6, 8, 10, 12, 14, 16, 18]}
# g_model = GridSearchCV(model, param_grid=params, scoring="f1", cv=5, n_jobs=-1, iid=False, verbose=0)
# g_model.fit(train_x, train_y)
# print(g_model.grid_scores_)
# print(g_model.best_score_)
# print(g_model.best_params_)
# log.info("Best parameter is {}, with score {}".format(g_model.best_params_, g_model.best_score_))
model.fit(train_x, train_y)
pred = model.predict(test_x)
test_uid = test.iloc[:, 0]
result = pd.DataFrame(columns=["uid", "label"])
result["uid"] = test_uid
result["label"] = pred
result.to_csv("result.csv", index=False)
model.booster().save_model("2.model")
class classifier:
def __init__(self):
self.model = XGBClassifier()
self.progress = 0
def para_tuning(
self,
X,
y,
para,
grid,
seed=0,
verbose=False
): # verbose = 1 for tuning log, verbose = 2 for plotting, verbose = 3 for both
# determine which to parameter to tune this time
if para == '':
return None
elif para == 'learning_rate':
param_grid = dict(learning_rate=grid) # [0,0.1]
elif para == 'max_depth':
param_grid = dict(max_depth=grid) # int
elif para == 'min_child_weight':
param_grid = dict(min_child_weight=grid) # [0,1]
elif para == 'gamma':
param_grid = dict(gamma=grid) # [0,1]
elif para == 'max_delta_step':
param_grid = dict(max_delta_step=grid) # int
elif para == 'colsample_bytree':
param_grid = dict(colsample_bytree=grid) # [0,1]
elif para == 'reg_alpha':
param_grid = dict(reg_alpha=grid) # [0,1]
elif para == 'reg_lambda':
param_grid = dict(reg_lambda=grid) # [0,1]
else:
print('WRONG PARAMETER.')
return None
kfold = StratifiedKFold(n_splits=8, shuffle=True, random_state=seed)
grid_search = GridSearchCV(self.model,
param_grid,
scoring='accuracy',
n_jobs=-1,
cv=kfold)
grid_result = grid_search.fit(X, y)
# summarize results
means = grid_result.cv_results_['mean_test_score']
stds = grid_result.cv_results_['std_test_score']
params = grid_result.cv_results_['params']
if verbose == 1 or verbose == 3:
for mean, stdev, param in zip(means, stds, params):
print('{:.4f} ({:.4f}) WITH: {} = {}'.format(
mean, stdev, para,
list(param.values())[0]))
print('-' * 63)
self.progress += 1
progress = int(self.progress / 7 * 100)
progress_bar = int(self.progress / 7 * 58)
print('\r' + '█' * progress_bar + ' ' * (58 - progress_bar) +
' {:>3}%'.format(progress),
end='')
if verbose == 2 or verbose == 3:
# plot
plt.close()
plt.figure(figsize=(20, 10))
plt.errorbar(grid, means, yerr=stds)
plt.title('XGBoost {} Tuning'.format(para))
plt.xlabel(para)
plt.ylabel('accuracy')
plt.show()
return list(grid_result.best_params_.values())[0]
def tune(self, X, y, verbose=False, seed=0):
self.model.seed = seed
# fit model no training data
print('-' * 63)
print('AUTO TUNING ON TRAINING DATASET.')
self.model.n_estimators = 1024
self.model.subsample = 0.6
self.model.learning_rate = 0.01
self.model.max_depth = self.para_tuning(X, y, 'max_depth',
[2, 4, 6, 8], seed, verbose)
self.model.min_child_weight = self.para_tuning(X, y,
'min_child_weight',
[4, 8, 12, 16], seed,
verbose)
self.model.gamma = self.para_tuning(
X, y, 'gamma', [0, 0.1, 0.2, 0.4, 0.8, 1.6, 3.2, 6.4, 12.8], seed,
verbose)
self.model.max_delta_step = self.para_tuning(X, y, 'max_delta_step',
[0, 1, 2, 4], seed,
verbose)
self.model.colsample_bytree = self.para_tuning(X, y,
'colsample_bytree',
[0.5, 0.6, 0.7], seed,
verbose)
self.model.reg_alpha = self.para_tuning(X, y, 'reg_alpha',
[0, 0.001, 0.01, 0.1, 10, 100],
seed, verbose)
self.model.reg_lambda = self.para_tuning(
X, y, 'reg_lambda', [0, 0.001, 0.01, 0.1, 10, 100], seed, verbose)
self.model.learning_rate /= 2
sleep(3)
print('\rAUTO TUNING FINISHED.' + ' ' * 42)
print('-' * 63)
if input('MODEL REVIEWING? (Y/N) ') == 'Y':
print(self.model)
def train(self, data, early_stopping_rounds=None, verbose=True, seed=0):
X_train, y_train = data.train[0], data.train[1]
X_test, y_test = data.test[0], data.test[1]
# tune paramters using trainging dataset
self.tune(X_train, y_train, seed=seed)
print('-' * 63)
# train the model with optimized parameters
print('MODEL TRAINING.')
metric = ['error', 'logloss', 'auc']
# self.model.min_child_weight = 4
self.model.fit(X_train,
y_train,
eval_metric=metric,
eval_set=[(X_train, y_train), (X_test, y_test)],
early_stopping_rounds=early_stopping_rounds,
verbose=False)
# make predictions for train data
y_pred = self.model.predict(X_train)
predictions = [round(value) for value in y_pred]
# evaluate predictions
accuracy = accuracy_score(y_train, predictions)
print('TRAINING FINISHED.')
print('ACCURACY TRAINING: {:.2f}%'.format(accuracy * 100))
# make predictions for test data
y_pred = self.model.predict(X_test)
predictions = [round(value) for value in y_pred]
# evaluate predictions
accuracy = accuracy_score(y_test, predictions)
print('ACCURACY TESTING: {:.2f}%'.format(accuracy * 100))
if verbose is True:
try:
# plot boosting results
results = self.model.evals_result()
epochs = len(results['validation_0'][metric[0]])
x_axis = range(0, epochs)
plt.style.use('ggplot')
plt.rcParams['font.size'] = 8
plt.figure(figsize=(20, 10))
i = 0
for m in metric:
ax = plt.subplot2grid((len(metric), 2), (i, 0))
i += 1
ax.plot(x_axis, results['validation_0'][m], label='Train')
ax.plot(x_axis, results['validation_1'][m], label='Test')
ax.legend()
ax.set_ylabel(m)
# plot feature importances
features = data.features
mapFeat = dict(
zip(['f' + str(i) for i in range(len(features))],
features))
imp = pd.Series(self.model.booster().get_fscore())
imp.index = imp.reset_index()['index'].map(mapFeat)
ax = plt.subplot2grid((len(metric), 2), (0, 1),
rowspan=len(metric))
imp.sort_values().plot(kind='barh')
ax.set_ylabel('importance')
plt.show()
except:
print('PLOTTING ERROR.')
class XGBoostModel:
def __init__(self, use_rfc=True):
self.use_rfc = use_rfc
if self.use_rfc:
# Instantiate Random Forest Classifier
self.rfc = RFCModel()
self.rfc.unpickle()
def load_train_data(self):
self.df, y, _ = clean_df('data/data.json', training=True)
if self.use_rfc:
# Include results from random forest classifier as new column
rfc_probs = self.rfc.predict_proba_all()
self.df['rfc_proba'] = rfc_probs
X = self.df.values
self.features = self.df.columns
self.X_train, self.X_test, self.y_train, self.y_test = train_test_split(
X, y, test_size=0.20, stratify=y, random_state=42)
def load_test_data(self):
self.df, _, oid = clean_df('data/data_point.json', training=False)
if self.use_rfc:
# Include results from random forest classifier as new column
rfc_probs = self.rfc.predict_proba('data/data_point.json')
self.df['rfc_proba'] = rfc_probs
return self.df.values, oid
def load_one(self, one_json):
# with open('one.json', 'w') as f:
# temp = '[' + one_json + ']'
# f.write(temp)
self.df, _, oid = clean_df('[' + one_json + ']', training=False)
if self.use_rfc:
# Include results from random forest classifier as new column
rfc_probs = self.rfc.predict_proba('data/data_point.json')
self.df['rfc_proba'] = rfc_probs
return self.df.values, oid
def fit(self):
self.model = XGBClassifier(max_depth=8,\
# reg_alpha=.8,\
n_estimators=200,\
scale_pos_weight=10.13,\
learning_rate=0.1)
self.model.fit(self.X_train, self.y_train)
@property
def feature_importances_(self):
#I couldn't call the master class, so just copy-n-pasted
#See https://github.com/dmlc/xgboost/commit/dd477ac903eb6f658d6fb2984763c3f8a4516389#diff-2c197a11c1b576e821f5942be9eab74c
b = self.model.booster()
fs = b.get_fscore()
all_features = [fs.get(f, 0.) for f in b.feature_names]
all_features = np.array(all_features, dtype=np.float32)
return all_features / all_features.sum()
def plot_features(self, save_img_dir=None, img_name_prefix='', ext='svg'):
'''
use ext='svg' for web!
add save_file_dir location to save images
save_file_dir has NO trailing slash!
eg 'static/images'
to keep multiple images saved add prefix string
prefix will be added to image file name
'''
# this is needed to fix lable clipping in saved files
from matplotlib import rcParams
rcParams.update({'figure.autolayout': True})
#severly modified from https://gist.github.com/light94/6c42df29f3232ae31e52
b = self.model.booster()
fs = b.get_fscore()
#print('feature...')
#print(b.feature_names)
#all_features = {f:fs.get(f, 0.) for f in b.feature_names}
#need to add real feature names
all_features = {
self.features[i]: float(fs.get('f' + str(i), 0.))
for i in range(len(b.feature_names))
}
importance = sorted(all_features.items(), key=itemgetter(1))
ff = pd.DataFrame(importance, columns=['feature', 'fscore'])
ff['fscore'] = ff['fscore'] / ff['fscore'].sum()
#"plot 1"
ax = ff.fscore.plot(xticks=ff.index, rot=65)
ax.set_xticklabels(ff.feature)
plt.title('XGBoost F-scores by feature')
if save_img_dir is not None:
plt.savefig('{}/{}feature_fscores.{}'.format(
save_img_dir, img_name_prefix, ext))
plt.show()
#"plot 2"
ff.plot(kind='barh',
x='feature',
y='fscore',
legend=False,
figsize=(6, 10))
plt.title('XGBoost Feature Importance')
plt.xlabel('relative importance')
if save_img_dir is not None:
plt.savefig('{}/{}features_barh.{}'.format(save_img_dir,
img_name_prefix, ext))
plt.show()
plt.close()
def pickle(self):
_pickle(self.model, 'data/XGBoostModel.pkl')
def unpickle(self):
self.model = _unpickle('data/XGBoostModel.pkl')
def score(self):
y_pred = self.model.predict(self.X_test)
probs = self.model.predict_proba(self.X_test)[:, 1]
accuracy = accuracy_score(self.y_test, y_pred)
f1 = f1_score(self.y_test, y_pred)
print("Accuracy: %.2f%%" % (accuracy * 100.0))
print("f1: %.2f" % f1)
print('Confusion matrix')
print(np.array([['TN', 'FP'], ['FN', 'TP']]))
print(confusion_matrix(self.y_test, y_pred))
def predict(self, X):
return self.model.predict(X)
def predict_proba(self, X):
prob = self.model.predict_proba(X)
return prob[:, 1]
def MLdecTree (learnData, picpath, endpoint="I9_STR_EXH", delCol=["I9_STR_SAH","I9_SEQULAE", "I9_STR", "IX_CIRCULATORY"], corrValue=0.995, binary=True):
#reads in processed Data from other function
learnColumn=learnData.columns
#correlates all the nevt columns with the target columns and saves the columns with high corr in list
matching = [s for s in learnColumn if endpoint.lower() in s.lower()]
endpointofInterest = [s for s in matching if "nevt" in s.lower()]
corrDropCol=[]
for colName in learnData.columns:
#print(colName)
if "nevt" in colName.lower():
coreName=colName.split('_NEVT')[0]
for match in matching:
corrCo=learnData[match].corr(learnData[colName], method='spearman')
if (corrCo > corrValue) or (corrCo < -corrValue):
#spike_cols = [col for col in learnColumn if coreName in col]
corrDropCol.extend([colName, coreName+"_AGE"])
#setting the y for endpoint of interest
y = learnData[endpointofInterest[0]].copy().to_numpy()
y = y.astype(int)
#drop all columns which are medicly too close related to endpoint
mask_pattrn = '|'.join(delCol)
if mask_pattrn:
learnData1 = learnData[learnData.columns.drop(list(learnData.filter(regex=mask_pattrn)))]
#deletes all strongly corr columns
corrDropCol=list(set(corrDropCol)-set(matching))
mask_pattrn = '|'.join(corrDropCol)
if mask_pattrn:
learnData1 = learnData1[learnData1.columns.drop(list(learnData1.filter(regex=mask_pattrn)))]
#Splitting dependent and independent Variable y=result
mask_pattrn = '|'.join(matching)
if mask_pattrn:
X = learnData1[learnData1.columns.drop(list(learnData1.filter(regex=mask_pattrn)))]
else:
X=learnData1
#splitting Data in train and test Data set
y=pd.Series(preprocessing.LabelEncoder().fit_transform(np.array(y)))
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42, test_size=0.2)
# ka=pd.Series(y.unique()).sort_values()
# , use_label_encoder=False
#fitting xgbTree
# clf_xgb= xgb.XGBClassifier(use_label_encoder=False) objective="multi:softmax", num_class = len(y.unique()))
# clf_xgb.fit(X_train, y_train, eval_metric="merror", eval_set=[(X_test, y_test)])
if binary is True:
#to be modified: gamma, n_jobs(threads)
#normal weight: scale_pos_weight= (y != 0).sum()/(y == 0).sum()
model = XGBClassifier(base_score=0.5, booster="gbtree", colsample_bylevel=1, colsample_bynode=1,
colsample_bytree=1, gamma=0.25, learning_rate=0.1, max_delta_step=0,
max_depth=6, min_child_weight=1, missing=None, n_estimators=100, n_jobs=1,
objective="binary:logistic", random_state=0, reg_alpha=0, reg_lambda=1,
scale_pos_weight=30, seed=None, subsample=1, verbosity=1)
#eval_metric:aucpr + auc + logloss
model.fit(X_train, y_train, verbose=True, eval_metric="auc")
else:
lc = LabelEncoder()
lc = lc.fit(y)
model = XGBClassifier(base_score=0.5, booster="gbtree", colsample_bylevel=1, colsample_bynode=1,
colsample_bytree=1, gamma=0, learning_rate=0.1, max_delta_step=0,
max_depth=6, min_child_weight=1, missing=None, n_estimators=100, n_jobs=1,
objective="reg:tweedie", random_state=0, reg_alpha=0, reg_lambda=1,
scale_pos_weight=(y == 0).sum()/(y != 0).sum(), seed=None, subsample=1, verbosity=1)
#eval_metric:rmse + tweedie-nloglik
model.fit(X_train, y_train, verbose=True, eval_metric="tweedie-nloglik")
#The accuracy of the model is calculated and printed
y_pred = model.predict(X_test)
predictions = [round(value) for value in y_pred]
accuracy = accuracy_score(y_test, predictions)
print("Accuracy: %.2f%%" % (accuracy * 100.0))
#Confusion plot (makes sense when the value is binary classified)
conf = plot_confusion_matrix(model,
X_test,
y_test,
display_labels=["Have no stroke", "Have a stroke"])
plt.savefig(picpath + '/confmatrix', format = "png")
#Code for printing out the xgb Tree calculated and make it pretty
bst = model.booster()
#for importance_type in ("weight","gain","cover","total_gain","total_cover"):
# print("%s: " % importance_type, bst.get_score(importance_type=importance_type))
#next two section is to make visual adjustments
node_params = {"shape": "box",
"style": "filled, rounded",
"fillcolor": "#78cbe"}
leaf_params= {"shape" : "box",
"style" : "filled",
"fillcolor" : "#e48038"}
#creates tree
image = xgb.to_graphviz(model, num_trees=0, size="10,10",
condition_node_params=node_params,
leaf_node_params=leaf_params)
#Set a different dpi (work only if format == 'png')
image.graph_attr = {'dpi':'400'}
#Saving the tree where the code is saved
image.render(picpath + '/modellbild1', format = "png")
return accuracy, model, corrDropCol
