🪴AutoGluonSelectML.py

Trains a model using AutoGluon on provided data path and returns feature importance and model leaderboard.

---

Parameters

• gene_data_path (str):

  • Path to the gene expression data CSV file.

  • For example: '../data/gene_tpm.csv'

• class_data_path (str):

  • Path to the class data CSV file.

  • For example: '../data/tumor_class.csv'

• label_column (str):

  • Name of the column in the dataset that is the target label for prediction.

• test_size (float):

  • Proportion of the data to be used as the test set.

• threshold (float):

  • The threshold used to filter out rows based on the proportion of non-zero values.

• hyperparameters (dict, optional):

  • Dictionary of hyperparameters for the models.

  • For example: {'GBM': {}, 'RF': {}}

• random_feature (int, optional):

  • The number of random feature to select. If None, no random feature selection is performed.

  • Default is None.

• num_bag_folds (int, optional)

  • Please note: This parameter annotation source can be referred to the documentation link in References.

  • Number of folds used for bagging of models. When num_bag_folds = k, training time is roughly increased by a factor of k (set = 0 to disable bagging). Disabled by default (0), but we recommend values between 5-10 to maximize predictive performance. Increasing num_bag_folds will result in models with lower bias but that are more prone to overfitting. num_bag_folds = 1 is an invalid value, and will raise a ValueError. Values > 10 may produce diminishing returns, and can even harm overall results due to overfitting. To further improve predictions, avoid increasing num_bag_folds much beyond 10 and instead increase num_bag_sets.

  • default = None

• num_stack_levels (int, optional)

  • Please note: This parameter annotation source can be referred to the documentation link in References.

  • Number of stacking levels to use in stack ensemble. Roughly increases model training time by factor of num_stack_levels+1 (set = 0 to disable stack ensembling). Disabled by default (0), but we recommend values between 1-3 to maximize predictive performance. To prevent overfitting, num_bag_folds >= 2 must also be set or else a ValueError will be raised.

  • default = None

• time_limit (int, optional):

  • Time limit for training in seconds.

  • Default is 120.

• random_state (int, optional):

  • The seed used by the random number generator.

  • Default is 42.

---

Return

• importance (DataFrame):

  • DataFrame containing feature importance.

• leaderboard (DataFrame):

  • DataFrame containing model performance on the test data.

---

References

Scientific Publications

AutoGluon-Tabular: Robust and Accurate AutoML for Structured DataarXiv.org

Fast, Accurate, and Simple Models for Tabular Data via Augmented...arXiv.org

Multimodal AutoML on Structured Tables with Text FieldsOpenReview

Articles

Machine learning with AutoGluon, an open source AutoML library | Amazon Web ServicesAmazon Web Services

https://medium.com/@zhanghang0704/image-classification-on-kaggle-using-autogluon-fc896e74d7e8medium.com

Accurate image classification in 3 lines of code with AutoGluon (Medium, Feb 2020)

https://towardsdatascience.com/autogluon-deep-learning-automl-5cdb4e2388ectowardsdatascience.com

AutoGluon overview & example applications (Towards Data Science, Dec 2019)

Documentation

AutoGluon Predictors — AutoGluon Documentation 0.1.0 documentation

Usage

Autogluon_TimeLimit(gene_data_path='../data/gene_tpm.csv', class_data_path='../data/tumor_class.csv', label_column='sex', test_size=0.3, threshold=0.9, hyperparameters={'GBM': {}, 'RF': {}},random_feature=none, num_bag_folds=None, num_stack_levels=None, time_limit=120, random_state=42)

---