Model Fitting#

単一モデルの Fitting#

多くのモデル候補にて学習が実行される。

INFO: Fitting 11 L1 models ...
INFO: Hyperparameter tuning model: LightGBMXT_BAG_L1 ...
INFO: 	Ran out of time, early stopping on iteration 3. Best iteration is:
	[3]	train_set's binary_logloss: 0.539284	valid_set's binary_logloss: 0.54396
INFO: 	Time limit exceeded
INFO: Fitted model: LightGBMXT_BAG_L1/T0 ...
INFO: 	0.8281	 = Validation score   (roc_auc)
INFO: 	53.13s	 = Training   runtime
INFO: 	0.34s	 = Validation runtime
...(中略)...
INFO: Fitting model: LightGBMLarge_BAG_L1 ... Training model for up to 1504.03s of the 1504.03s of remaining time.
INFO: 	0.8286	 = Validation score   (roc_auc)
INFO: 	228.19s	 = Training   runtime
INFO: 	0.09s	 = Validation runtime

アンサンブル#

続いて、ここまで fit されたモデル群(L1)のアンサンブル(L2)が行われている。

INFO: Completed 1/2 k-fold bagging repeats ...
INFO: Fitting model: WeightedEnsemble_L2 ... Training model for up to 360.0s of the 1279.97s of remaining time.
INFO: 	0.8473	 = Validation score   (roc_auc)
INFO: 	2.66s	 = Training   runtime
INFO: 	0.0s	 = Validation runtime
INFO: AutoGluon training complete, total runtime = 2322.71s ...
INFO: TabularPredictor saved. To load, use: predictor = TabularPredictor.load("ag_models/")

使用したメモリリソース情報#

peak memory: 1156.41 MiB, increment: 743.93 MiB
CPU times: user 4h 34min 17s, sys: 12min 12s, total: 4h 46min 29s
Wall time: 38min 43s

学習で消費したメモリ量を peak memory から知ることができる。このメモリ量が TD AutoML 実行時に指定した task memory を超過していないかをチェックする必要がある。

モデルの決定