def train_level1(self, select_param=False):
self.l1_param = {
"bst:max_depth": 10,
"bst:min_child_weight": 4,
"bst:subsample": 0.5,
"bst:colsample_bytree": 0.8,
"bst:eta": 0.05,
}
other = {
"silent": 1,
"objective": "multi:softprob",
"num_class": 9,
"nthread": 4,
"eval_metric": "mlogloss",
"seed": 0,
}
other["num_class"] = len(self.l1_lbl_enc.classes_)
self.l1_num_round = 400
if select_param:
results = otto_utils.select_xgb_params(self.l1_train, self.l1_labels, depth_range=(7, 12))
self.l1_results = results
print results[0]
self.l1_num_round = results[0][1]
param = results[0][2]
dtrain = xgb.DMatrix(self.l1_train, label=self.l1_labels)
full_param = other.copy()
full_param.update(self.l1_param)
bst, loss, ntree = xgb.train(full_param, dtrain, self.l1_num_round, [(dtrain, "train")])
self.l1_model = bst
pass
def train_level2(self, select_param=False):
self.l2_models = []
self.l2_params = []
self.l2_num_rounds = []
self.l2_results = []
for i in xrange(len(self.cls_sets)):
param = {
"bst:max_depth": 10,
"bst:min_child_weight": 4,
"bst:subsample": 0.5,
"bst:colsample_bytree": 0.8,
"bst:eta": 0.05,
}
other = {
"silent": 1,
"objective": "multi:softprob",
"num_class": 9,
"nthread": 4,
"eval_metric": "mlogloss",
"seed": 0,
}
other["num_class"] = len(self.cls_sets[i])
num_round = 400
if select_param:
results = otto_utils.select_xgb_params(self.l2_train[i], self.l2_labels[i], depth_range=(8, 13))
self.l2_results.append(results)
print results[0]
num_round = results[0][1]
param = results[0][2]
dtrain = xgb.DMatrix(self.l2_train[i], self.l2_labels[i])
self.l2_params.append(param)
self.l2_num_rounds.append(num_round)
full_param = other.copy()
full_param.update(param)
bst, loss, ntree = xgb.train(full_param, dtrain, num_round, [(dtrain, "train")])
self.l2_models.append(bst)
pass
