いろんなAutoMLを試してみてます。 こちらはblueqat cloudの有料版で利用できました。pythonのバージョンの関係で。 なんかいろいろありますが、評判がよさそうなAutoGluon使ってみます。pythonのバージョンを3.6で実行しました。 <pre class="ql-syntax" spellcheck="false">!python3 -m pip install -U pip
!python3 -m pip install -U setuptools wheel
!python3 -m pip install -U &#34;mxnet&lt;2.0.0&#34;
!python3 -m pip install autogluon
</pre> ツールを読み込みます。なんかうまく読めなかったので、こちらの記事を参照しながら進めました。<a href="https://qiita.com/dyamaguc/items/dded739f35e59a6491c8" rel="noopener noreferrer" target="_blank">https://qiita.com/dyamaguc/items/dded739f35e59a6491c8</a> <pre class="ql-syntax" spellcheck="false">from autogluon import TabularPrediction
</pre> データは5000にしてみました。<pre class="ql-syntax" spellcheck="false">train_data = TabularPrediction.Dataset(file_path=&#39;https://autogluon.s3.amazonaws.com/datasets/Inc/train.csv&#39;)
subsample_size = 5000
train_data = train_data.sample(n=subsample_size, random_state=0)
train_data.head()
</pre> 目的変数をclassにします。classは2値になってます。<pre class="ql-syntax" spellcheck="false">label = &#39;class&#39;
print(&#34;Summary of class variable: \n&#34;, train_data[label].describe())
</pre> さっそく時間を測定して、予測をしてみます。<pre class="ql-syntax" spellcheck="false">%%time
save_path = &#39;agModels-predictClass&#39; # specifies folder to store trained models
predictor = TabularPrediction.fit(train_data=train_data, label=label, output_directory=dir)
</pre> 学習結果を見てみます。<pre class="ql-syntax" spellcheck="false">results = predictor.fit_summary()
</pre> 学習結果は、スコアの良かった順にモデルが並んでいます。すごい簡単に使えました。<pre class="ql-syntax" spellcheck="false">*** Summary of fit() ***
Estimated performance of each model:
 model score_val pred_time_val fit_time pred_time_val_marginal fit_time_marginal stack_level can_infer fit_order
0 weighted_ensemble_k0_l1 0.878 0.280509 5.315961 0.002607 0.637513 1 True 12
1 CatboostClassifier 0.872 0.021003 3.832444 0.021003 3.832444 0 True 9
2 LightGBMClassifierXT 0.870 0.024198 0.412143 0.024198 0.412143 0 True 8
3 LightGBMClassifier 0.862 0.023735 0.412788 0.023735 0.412788 0 True 7
4 NeuralNetClassifier 0.862 0.184480 21.246298 0.184480 21.246298 0 True 10
5 LightGBMClassifierCustom 0.856 0.024570 0.710194 0.024570 0.710194 0 True 11
6 ExtraTreesClassifierGini 0.852 0.111771 0.884064 0.111771 0.884064 0 True 3
7 RandomForestClassifierGini 0.850 0.111832 1.104524 0.111832 1.104524 0 True 1
8 RandomForestClassifierEntr 0.846 0.111633 1.388850 0.111633 1.388850 0 True 2
9 ExtraTreesClassifierEntr 0.836 0.111355 0.983722 0.111355 0.983722 0 True 4
10 KNeighborsClassifierUnif 0.750 0.104517 0.011417 0.104517 0.011417 0 True 5
11 KNeighborsClassifierDist 0.732 0.104449 0.009657 0.104449 0.009657 0 True 6
Number of models trained: 12
Types of models trained:
{&#39;WeightedEnsembleModel&#39;, &#39;TabularNeuralNetModel&#39;, &#39;CatboostModel&#39;, &#39;RFModel&#39;, &#39;KNNModel&#39;, &#39;XTModel&#39;, &#39;LGBModel&#39;}
Bagging used: False 
Stack-ensembling used: False 
Hyperparameter-tuning used: False 
User-specified hyperparameters:
{&#39;default&#39;: {&#39;NN&#39;: [{}], &#39;GBM&#39;: [{}, {&#39;extra_trees&#39;: True, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;XT&#39;}}], &#39;CAT&#39;: [{}], &#39;RF&#39;: [{&#39;criterion&#39;: &#39;gini&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Gini&#39;, &#39;problem_types&#39;: [&#39;binary&#39;, &#39;multiclass&#39;]}}, {&#39;criterion&#39;: &#39;entropy&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Entr&#39;, &#39;problem_types&#39;: [&#39;binary&#39;, &#39;multiclass&#39;]}}], &#39;XT&#39;: [{&#39;criterion&#39;: &#39;gini&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Gini&#39;, &#39;problem_types&#39;: [&#39;binary&#39;, &#39;multiclass&#39;]}}, {&#39;criterion&#39;: &#39;entropy&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Entr&#39;, &#39;problem_types&#39;: [&#39;binary&#39;, &#39;multiclass&#39;]}}], &#39;KNN&#39;: [{&#39;weights&#39;: &#39;uniform&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Unif&#39;}}, {&#39;weights&#39;: &#39;distance&#39;, &#39;AG_args&#39;: {&#39;name_suffix&#39;: &#39;Dist&#39;}}], &#39;custom&#39;: [{&#39;num_boost_round&#39;: 10000, &#39;num_threads&#39;: -1, &#39;objective&#39;: &#39;binary&#39;, &#39;verbose&#39;: -1, &#39;boosting_type&#39;: &#39;gbdt&#39;, &#39;learning_rate&#39;: 0.03, &#39;num_leaves&#39;: 128, &#39;feature_fraction&#39;: 0.9, &#39;min_data_in_leaf&#39;: 5, &#39;two_round&#39;: True, &#39;seed_value&#39;: 0, &#39;AG_args&#39;: {&#39;model_type&#39;: &#39;GBM&#39;, &#39;name_suffix&#39;: &#39;Custom&#39;, &#39;disable_in_hpo&#39;: True}}]}}
Feature Metadata (Processed):
(raw dtype, special dtypes):
(&#39;category&#39;, []) : 8 | [&#39;workclass&#39;, &#39;education&#39;, &#39;marital-status&#39;, &#39;occupation&#39;, &#39;relationship&#39;, ...]
(&#39;int&#39;, []) : 6 | [&#39;age&#39;, &#39;fnlwgt&#39;, &#39;education-num&#39;, &#39;capital-gain&#39;, &#39;capital-loss&#39;, ...]
Plot summary of models saved to file: agModels-predictClass/SummaryOfModels.html
*** End of fit() summary ***
</pre> 前の記事では、Auto Sklearnや今回Auto Gluonを使ってみました。簡単にいろんなモデルを比較できるので簡単ですね。データ業界は競争が激しいのだなと感じました。以上ですが、もっといろんなデータに関するテクニックを学んで実装をしてみたいと思いました。

AutoGluon試してみた

Yuichiro Minato