Adding a custom metric to AutoGluon¶
Tip: If you are new to AutoGluon, review Predicting Columns in a Table - Quick Start to learn the basics of the AutoGluon API.
This tutorial describes how to add a custom evaluation metric to AutoGluon that is used to inform validation scores, model ensembling, hyperparameter tuning, and more.
In this example, we show a variety of evaluation metrics and how to convert them to an AutoGluon Scorer, which can then be passed to AutoGluon models and predictors.
First, we will randomly generate 10 ground truth labels and predictions and show how to calculate metric scores from them.
import numpy as np
y_true = np.random.randint(low=0, high=2, size=10)
y_pred = np.random.randint(low=0, high=2, size=10)
print(f'y_true: {y_true}')
print(f'y_pred: {y_pred}')
y_true: [1 1 0 1 1 1 0 0 1 0]
y_pred: [1 0 0 1 1 0 1 1 1 1]
Ensuring Metric is Serializable¶
Custom metrics must be defined in a separate Python file and imported so that they can be pickled (Python’s serialization protocol).
If a customer metric is not picklable, AutoGluon will crash during fit when trying to parallelize model training with Ray.
In the below example, you would want to create a new python file such as my_metrics.py with ag_accuracy_scorer defined in it,
and then use it via from my_metrics import ag_accuracy_scorer.
If your metric is not serializable, you will get many errors similar to: _pickle.PicklingError: Can't pickle. Refer to https://github.com/autogluon/autogluon/issues/1637 for an example.
The custom metrics in this tutorial are not serializable for ease of demonstration. If the best_quality preset was used, calls to fit() would crash.
Custom Accuracy Metric¶
We will start by creating a customer accuracy metric. A prediction is correct if the predicted value is the same as the true value, otherwise it is wrong.
import sklearn.metrics
sklearn.metrics.accuracy_score(y_true, y_pred)
0.5
There are a variety of limitations with the above logic. For example, without outside knowledge of the metric it is unknown:
What the optimal value is (1)
If higher values are better (True)
If the metric requires prediction labels or probabilities (labels)
Now, let’s convert this evaluation metric to an AutoGluon Scorer to address these limitations.
We do this by calling autogluon.core.metrics.make_scorer.
from autogluon.core.metrics import make_scorer
ag_accuracy_scorer = make_scorer(name='accuracy',
score_func=sklearn.metrics.accuracy_score,
optimum=1,
greater_is_better=True)
When creating the Scorer, we need to specify a name for the Scorer. This does not need to be any particular value but is used when printing information about the Scorer during training.
Next, we specify the score_func. This is the function we want to wrap, in this case, sklearn’s accuracy_score function.
We then need to specify the optimum value.
This is necessary when calculating error (also known as regret) as opposed to score.
error is defined as sign * optimum - score, where sign=1 if greater_is_better=True, else sign=-1.
It is also useful to identify when a score is optimal and cannot be improved.
Because the best possible value from sklearn.metrics.accuracy_score is 1, we specify optimum=1.
Finally, we need to specify greater_is_better. In this case, greater_is_better=True
because the best value returned is 1, and the worst value returned is less than 1 (0).
It is very important to set this value correctly,
otherwise AutoGluon will try to optimize for the worst model instead of the best.
Advanced Note: optimum must correspond to the optimal value
from the original metric callable (in this case sklearn.metrics.accuracy_score).
Hypothetically, if a metric callable was greater_is_better=False with an optimal value of -2,
you should specify optimum=-2, greater_is_better=False.
In this case, if raw_metric_value=-0.5
then Scorer would return score=0.5 to enforce higher_is_better (score = sign * raw_metric_value).
Scorer’s error would be error=1.5 because sign (-1) * optimum (-2) - score (0.5) = 1.5
Once created, the AutoGluon Scorer can be called in the same fashion as the original metric to compute score.
# score
ag_accuracy_scorer(y_true, y_pred)
0.5
Alternatively, .score is an alias to the above callable for convenience:
ag_accuracy_scorer.score(y_true, y_pred)
0.5
To get the error instead of score:
# error, error=sign*optimum-score -> error=1*1-score -> error=1-score
ag_accuracy_scorer.error(y_true, y_pred)
# Can also convert score to error:
# score = ag_accuracy_scorer(y_true, y_pred)
# error = ag_accuracy_scorer.convert_score_to_error(score)
0.5
Note that score is in higher_is_better format, while error is in lower_is_better format.
An error of 0 corresponds to a perfect prediction.
Custom Mean Squared Error Metric¶
Next, let’s show examples of how to convert regression metrics into Scorers.
First we generate random ground truth labels and their predictions, however this time they are floats instead of integers.
y_true = np.random.rand(10)
y_pred = np.random.rand(10)
print(f'y_true: {y_true}')
print(f'y_pred: {y_pred}')
y_true: [0.57172623 0.93558456 0.02921353 0.11810811 0.16160864 0.0316515
0.83481643 0.87914478 0.5769894 0.43549493]
y_pred: [0.50280096 0.04173631 0.48941556 0.532765 0.74676418 0.3603305
0.0854697 0.85096748 0.28820601 0.6171923 ]
A common regression metric is Mean Squared Error:
sklearn.metrics.mean_squared_error(y_true, y_pred)
0.23166027866789132
ag_mean_squared_error_scorer = make_scorer(name='mean_squared_error',
score_func=sklearn.metrics.mean_squared_error,
optimum=0,
greater_is_better=False)
In this case, optimum=0 because this is an error metric.
Additionally, greater_is_better=False because sklearn reports error as positive values, and the lower the value is, the better.
A very important point about AutoGluon Scorers is that internally,
they will always report scores in greater_is_better=True form.
This means if the original metric was greater_is_better=False, AutoGluon’s Scorer will flip the value.
Therefore, score will be represented as a negative value.
This is done to ensure consistency between different metrics.
# score
ag_mean_squared_error_scorer(y_true, y_pred)
-0.23166027866789132
# error, error=sign*optimum-score -> error=-1*0-score -> error=-score
ag_mean_squared_error_scorer.error(y_true, y_pred)
0.23166027866789132
We can also specify metrics outside of sklearn. For example, below is a minimal implementation of mean squared error:
def mse_func(y_true: np.ndarray, y_pred: np.ndarray) -> float:
return ((y_true - y_pred) ** 2).mean()
mse_func(y_true, y_pred)
0.23166027866789132
All that is required is that the function take two arguments: y_true, and y_pred (or y_pred_proba), as numpy arrays, and return a float value.
With the same code as before, we can create an AutoGluon Scorer.
ag_mean_squared_error_custom_scorer = make_scorer(name='mean_squared_error',
score_func=mse_func,
optimum=0,
greater_is_better=False)
ag_mean_squared_error_custom_scorer(y_true, y_pred)
-0.23166027866789132
Custom ROC AUC Metric¶
Here we show an example of a thresholding metric, roc_auc. A thresholding metric cares about the relative ordering of predictions, but not their absolute values.
y_true = np.random.randint(low=0, high=2, size=10)
y_pred_proba = np.random.rand(10)
print(f'y_true: {y_true}')
print(f'y_pred_proba: {y_pred_proba}')
y_true: [1 1 0 0 1 0 0 0 0 0]
y_pred_proba: [0.87339138 0.83380033 0.35505658 0.35654049 0.74315981 0.96510012
0.47357165 0.55254438 0.72802271 0.34469704]
sklearn.metrics.roc_auc_score(y_true, y_pred_proba)
0.8571428571428572
We will need to specify needs_threshold=True in order for downstream models to properly use the metric.
# Score functions that need decision values
ag_roc_auc_scorer = make_scorer(name='roc_auc',
score_func=sklearn.metrics.roc_auc_score,
optimum=1,
greater_is_better=True,
needs_threshold=True)
ag_roc_auc_scorer(y_true, y_pred_proba)
0.8571428571428572
Using Custom Metrics in TabularPredictor¶
Now that we have created several custom Scorers, let’s use them for training and evaluating models.
For this tutorial, we will be using the Adult Income dataset.
from autogluon.tabular import TabularDataset
train_data = TabularDataset('https://autogluon.s3.amazonaws.com/datasets/Inc/train.csv') # can be local CSV file as well, returns Pandas DataFrame
test_data = TabularDataset('https://autogluon.s3.amazonaws.com/datasets/Inc/test.csv') # another Pandas DataFrame
label = 'class' # specifies which column we want to predict
train_data = train_data.sample(n=1000, random_state=0) # subsample dataset for faster demo
train_data.head(5)
| age | workclass | fnlwgt | education | education-num | marital-status | occupation | relationship | race | sex | capital-gain | capital-loss | hours-per-week | native-country | class | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 6118 | 51 | Private | 39264 | Some-college | 10 | Married-civ-spouse | Exec-managerial | Wife | White | Female | 0 | 0 | 40 | United-States | >50K |
| 23204 | 58 | Private | 51662 | 10th | 6 | Married-civ-spouse | Other-service | Wife | White | Female | 0 | 0 | 8 | United-States | <=50K |
| 29590 | 40 | Private | 326310 | Some-college | 10 | Married-civ-spouse | Craft-repair | Husband | White | Male | 0 | 0 | 44 | United-States | <=50K |
| 18116 | 37 | Private | 222450 | HS-grad | 9 | Never-married | Sales | Not-in-family | White | Male | 0 | 2339 | 40 | El-Salvador | <=50K |
| 33964 | 62 | Private | 109190 | Bachelors | 13 | Married-civ-spouse | Exec-managerial | Husband | White | Male | 15024 | 0 | 40 | United-States | >50K |
from autogluon.tabular import TabularPredictor
predictor = TabularPredictor(label=label).fit(train_data, hyperparameters='toy')
predictor.leaderboard(test_data)
No path specified. Models will be saved in: "AutogluonModels/ag-20240418_043719"
No presets specified! To achieve strong results with AutoGluon, it is recommended to use the available presets.
Recommended Presets (For more details refer to https://auto.gluon.ai/stable/tutorials/tabular/tabular-essentials.html#presets):
presets='best_quality' : Maximize accuracy. Default time_limit=3600.
presets='high_quality' : Strong accuracy with fast inference speed. Default time_limit=3600.
presets='good_quality' : Good accuracy with very fast inference speed. Default time_limit=3600.
presets='medium_quality' : Fast training time, ideal for initial prototyping.
Beginning AutoGluon training ...
AutoGluon will save models to "AutogluonModels/ag-20240418_043719"
=================== System Info ===================
AutoGluon Version: 1.1.0b20240418
Python Version: 3.10.12
Operating System: Linux
Platform Machine: x86_64
Platform Version: #1 SMP Tue Nov 30 00:17:50 UTC 2021
CPU Count: 8
Memory Avail: 28.85 GB / 30.96 GB (93.2%)
Disk Space Avail: 217.25 GB / 255.99 GB (84.9%)
===================================================
Train Data Rows: 1000
Train Data Columns: 14
Label Column: class
AutoGluon infers your prediction problem is: 'binary' (because only two unique label-values observed).
2 unique label values: [' >50K', ' <=50K']
If 'binary' is not the correct problem_type, please manually specify the problem_type parameter during predictor init (You may specify problem_type as one of: ['binary', 'multiclass', 'regression'])
Problem Type: binary
Preprocessing data ...
Selected class <--> label mapping: class 1 = >50K, class 0 = <=50K
Note: For your binary classification, AutoGluon arbitrarily selected which label-value represents positive ( >50K) vs negative ( <=50K) class.
To explicitly set the positive_class, either rename classes to 1 and 0, or specify positive_class in Predictor init.
Using Feature Generators to preprocess the data ...
Fitting AutoMLPipelineFeatureGenerator...
Available Memory: 29540.36 MB
Train Data (Original) Memory Usage: 0.56 MB (0.0% of available memory)
Inferring data type of each feature based on column values. Set feature_metadata_in to manually specify special dtypes of the features.
Stage 1 Generators:
Fitting AsTypeFeatureGenerator...
Note: Converting 1 features to boolean dtype as they only contain 2 unique values.
Stage 2 Generators:
Fitting FillNaFeatureGenerator...
Stage 3 Generators:
Fitting IdentityFeatureGenerator...
Fitting CategoryFeatureGenerator...
Fitting CategoryMemoryMinimizeFeatureGenerator...
Stage 4 Generators:
Fitting DropUniqueFeatureGenerator...
Stage 5 Generators:
Fitting DropDuplicatesFeatureGenerator...
Types of features in original data (raw dtype, special dtypes):
('int', []) : 6 | ['age', 'fnlwgt', 'education-num', 'capital-gain', 'capital-loss', ...]
('object', []) : 8 | ['workclass', 'education', 'marital-status', 'occupation', 'relationship', ...]
Types of features in processed data (raw dtype, special dtypes):
('category', []) : 7 | ['workclass', 'education', 'marital-status', 'occupation', 'relationship', ...]
('int', []) : 6 | ['age', 'fnlwgt', 'education-num', 'capital-gain', 'capital-loss', ...]
('int', ['bool']) : 1 | ['sex']
0.1s = Fit runtime
14 features in original data used to generate 14 features in processed data.
Train Data (Processed) Memory Usage: 0.06 MB (0.0% of available memory)
Data preprocessing and feature engineering runtime = 0.11s ...
AutoGluon will gauge predictive performance using evaluation metric: 'accuracy'
To change this, specify the eval_metric parameter of Predictor()
Automatically generating train/validation split with holdout_frac=0.2, Train Rows: 800, Val Rows: 200
User-specified model hyperparameters to be fit:
{
'NN_TORCH': {'num_epochs': 5},
'GBM': {'num_boost_round': 10},
'CAT': {'iterations': 10},
'XGB': {'n_estimators': 10},
}
Fitting 4 L1 models ...
Fitting model: LightGBM ...
0.77 = Validation score (accuracy)
0.25s = Training runtime
0.01s = Validation runtime
Fitting model: CatBoost ...
0.86 = Validation score (accuracy)
0.19s = Training runtime
0.01s = Validation runtime
Fitting model: XGBoost ...
0.845 = Validation score (accuracy)
0.11s = Training runtime
0.01s = Validation runtime
Fitting model: NeuralNetTorch ...
0.84 = Validation score (accuracy)
3.74s = Training runtime
0.01s = Validation runtime
Fitting model: WeightedEnsemble_L2 ...
Ensemble Weights: {'CatBoost': 1.0}
0.86 = Validation score (accuracy)
0.05s = Training runtime
0.0s = Validation runtime
AutoGluon training complete, total runtime = 4.53s ... Best model: "WeightedEnsemble_L2"
TabularPredictor saved. To load, use: predictor = TabularPredictor.load("AutogluonModels/ag-20240418_043719")
| model | score_test | score_val | eval_metric | pred_time_test | pred_time_val | fit_time | pred_time_test_marginal | pred_time_val_marginal | fit_time_marginal | stack_level | can_infer | fit_order | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | CatBoost | 0.842768 | 0.860 | accuracy | 0.007167 | 0.006709 | 0.188969 | 0.007167 | 0.006709 | 0.188969 | 1 | True | 2 |
| 1 | WeightedEnsemble_L2 | 0.842768 | 0.860 | accuracy | 0.009115 | 0.007545 | 0.238240 | 0.001948 | 0.000837 | 0.049271 | 2 | True | 5 |
| 2 | XGBoost | 0.838162 | 0.845 | accuracy | 0.044245 | 0.008597 | 0.111236 | 0.044245 | 0.008597 | 0.111236 | 1 | True | 3 |
| 3 | NeuralNetTorch | 0.827209 | 0.840 | accuracy | 0.051908 | 0.012972 | 3.741137 | 0.051908 | 0.012972 | 3.741137 | 1 | True | 4 |
| 4 | LightGBM | 0.780940 | 0.770 | accuracy | 0.006747 | 0.005919 | 0.254553 | 0.006747 | 0.005919 | 0.254553 | 1 | True | 1 |
We can pass our custom metrics into predictor.leaderboard via the extra_metrics argument:
predictor.leaderboard(test_data, extra_metrics=[ag_roc_auc_scorer, ag_accuracy_scorer])
| model | score_test | roc_auc | accuracy | score_val | eval_metric | pred_time_test | pred_time_val | fit_time | pred_time_test_marginal | pred_time_val_marginal | fit_time_marginal | stack_level | can_infer | fit_order | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | CatBoost | 0.842768 | 0.863760 | 0.842768 | 0.860 | accuracy | 0.006668 | 0.006709 | 0.188969 | 0.006668 | 0.006709 | 0.188969 | 1 | True | 2 |
| 1 | WeightedEnsemble_L2 | 0.842768 | 0.863760 | 0.842768 | 0.860 | accuracy | 0.009115 | 0.007545 | 0.238240 | 0.002447 | 0.000837 | 0.049271 | 2 | True | 5 |
| 2 | XGBoost | 0.838162 | 0.889922 | 0.838162 | 0.845 | accuracy | 0.058511 | 0.008597 | 0.111236 | 0.058511 | 0.008597 | 0.111236 | 1 | True | 3 |
| 3 | NeuralNetTorch | 0.827209 | 0.878940 | 0.827209 | 0.840 | accuracy | 0.069338 | 0.012972 | 3.741137 | 0.069338 | 0.012972 | 3.741137 | 1 | True | 4 |
| 4 | LightGBM | 0.780940 | 0.861131 | 0.780940 | 0.770 | accuracy | 0.006737 | 0.005919 | 0.254553 | 0.006737 | 0.005919 | 0.254553 | 1 | True | 1 |
We can also pass our custom metric into the Predictor itself by specifying it during initialization via the eval_metric parameter:
predictor_custom = TabularPredictor(label=label, eval_metric=ag_roc_auc_scorer).fit(train_data, hyperparameters='toy')
predictor_custom.leaderboard(test_data)
No path specified. Models will be saved in: "AutogluonModels/ag-20240418_043724"
No presets specified! To achieve strong results with AutoGluon, it is recommended to use the available presets.
Recommended Presets (For more details refer to https://auto.gluon.ai/stable/tutorials/tabular/tabular-essentials.html#presets):
presets='best_quality' : Maximize accuracy. Default time_limit=3600.
presets='high_quality' : Strong accuracy with fast inference speed. Default time_limit=3600.
presets='good_quality' : Good accuracy with very fast inference speed. Default time_limit=3600.
presets='medium_quality' : Fast training time, ideal for initial prototyping.
Beginning AutoGluon training ...
AutoGluon will save models to "AutogluonModels/ag-20240418_043724"
=================== System Info ===================
AutoGluon Version: 1.1.0b20240418
Python Version: 3.10.12
Operating System: Linux
Platform Machine: x86_64
Platform Version: #1 SMP Tue Nov 30 00:17:50 UTC 2021
CPU Count: 8
Memory Avail: 28.06 GB / 30.96 GB (90.6%)
Disk Space Avail: 217.25 GB / 255.99 GB (84.9%)
===================================================
Train Data Rows: 1000
Train Data Columns: 14
Label Column: class
AutoGluon infers your prediction problem is: 'binary' (because only two unique label-values observed).
2 unique label values: [' >50K', ' <=50K']
If 'binary' is not the correct problem_type, please manually specify the problem_type parameter during predictor init (You may specify problem_type as one of: ['binary', 'multiclass', 'regression'])
Problem Type: binary
Preprocessing data ...
Selected class <--> label mapping: class 1 = >50K, class 0 = <=50K
Note: For your binary classification, AutoGluon arbitrarily selected which label-value represents positive ( >50K) vs negative ( <=50K) class.
To explicitly set the positive_class, either rename classes to 1 and 0, or specify positive_class in Predictor init.
Using Feature Generators to preprocess the data ...
Fitting AutoMLPipelineFeatureGenerator...
Available Memory: 28731.71 MB
Train Data (Original) Memory Usage: 0.56 MB (0.0% of available memory)
Inferring data type of each feature based on column values. Set feature_metadata_in to manually specify special dtypes of the features.
Stage 1 Generators:
Fitting AsTypeFeatureGenerator...
Note: Converting 1 features to boolean dtype as they only contain 2 unique values.
Stage 2 Generators:
Fitting FillNaFeatureGenerator...
Stage 3 Generators:
Fitting IdentityFeatureGenerator...
Fitting CategoryFeatureGenerator...
Fitting CategoryMemoryMinimizeFeatureGenerator...
Stage 4 Generators:
Fitting DropUniqueFeatureGenerator...
Stage 5 Generators:
Fitting DropDuplicatesFeatureGenerator...
Types of features in original data (raw dtype, special dtypes):
('int', []) : 6 | ['age', 'fnlwgt', 'education-num', 'capital-gain', 'capital-loss', ...]
('object', []) : 8 | ['workclass', 'education', 'marital-status', 'occupation', 'relationship', ...]
Types of features in processed data (raw dtype, special dtypes):
('category', []) : 7 | ['workclass', 'education', 'marital-status', 'occupation', 'relationship', ...]
('int', []) : 6 | ['age', 'fnlwgt', 'education-num', 'capital-gain', 'capital-loss', ...]
('int', ['bool']) : 1 | ['sex']
0.1s = Fit runtime
14 features in original data used to generate 14 features in processed data.
Train Data (Processed) Memory Usage: 0.06 MB (0.0% of available memory)
Data preprocessing and feature engineering runtime = 0.12s ...
AutoGluon will gauge predictive performance using evaluation metric: 'roc_auc'
This metric expects predicted probabilities rather than predicted class labels, so you'll need to use predict_proba() instead of predict()
To change this, specify the eval_metric parameter of Predictor()
Automatically generating train/validation split with holdout_frac=0.2, Train Rows: 800, Val Rows: 200
User-specified model hyperparameters to be fit:
{
'NN_TORCH': {'num_epochs': 5},
'GBM': {'num_boost_round': 10},
'CAT': {'iterations': 10},
'XGB': {'n_estimators': 10},
}
Fitting 4 L1 models ...
Fitting model: LightGBM ...
0.85 = Validation score (roc_auc)
0.17s = Training runtime
0.01s = Validation runtime
Fitting model: CatBoost ...
0.8693 = Validation score (roc_auc)
0.04s = Training runtime
0.01s = Validation runtime
Fitting model: XGBoost ...
0.8591 = Validation score (roc_auc)
0.16s = Training runtime
0.01s = Validation runtime
Fitting model: NeuralNetTorch ...
0.8535 = Validation score (roc_auc)
0.49s = Training runtime
0.01s = Validation runtime
Fitting model: WeightedEnsemble_L2 ...
Ensemble Weights: {'XGBoost': 0.333, 'LightGBM': 0.292, 'CatBoost': 0.292, 'NeuralNetTorch': 0.083}
0.8788 = Validation score (roc_auc)
0.18s = Training runtime
0.0s = Validation runtime
AutoGluon training complete, total runtime = 1.25s ... Best model: "WeightedEnsemble_L2"
TabularPredictor saved. To load, use: predictor = TabularPredictor.load("AutogluonModels/ag-20240418_043724")
| model | score_test | score_val | eval_metric | pred_time_test | pred_time_val | fit_time | pred_time_test_marginal | pred_time_val_marginal | fit_time_marginal | stack_level | can_infer | fit_order | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | WeightedEnsemble_L2 | 0.900198 | 0.878800 | roc_auc | 0.145154 | 0.040682 | 1.039392 | 0.004006 | 0.002000 | 0.182074 | 2 | True | 5 |
| 1 | XGBoost | 0.889922 | 0.859126 | roc_auc | 0.056100 | 0.009703 | 0.156073 | 0.056100 | 0.009703 | 0.156073 | 1 | True | 3 |
| 2 | CatBoost | 0.887425 | 0.869325 | roc_auc | 0.006760 | 0.009200 | 0.035183 | 0.006760 | 0.009200 | 0.035183 | 1 | True | 2 |
| 3 | NeuralNetTorch | 0.878940 | 0.853533 | roc_auc | 0.071225 | 0.013422 | 0.491132 | 0.071225 | 0.013422 | 0.491132 | 1 | True | 4 |
| 4 | LightGBM | 0.870968 | 0.849980 | roc_auc | 0.007063 | 0.006357 | 0.174930 | 0.007063 | 0.006357 | 0.174930 | 1 | True | 1 |
That’s all it takes to create and use custom metrics in AutoGluon!
If you create a custom metric, consider submitting a PR so that we can officially add it to AutoGluon!
For a tutorial on implementing custom models in AutoGluon, refer to Adding a custom model to AutoGluon.
For more tutorials, refer to Predicting Columns in a Table - Quick Start and Predicting Columns in a Table - In Depth.