Automated machine learning is the practice of automating the inefficient, iterative jobs of machine learning model development. It is also stated as automated ML or AutoML. It permits data scientists, specialists, and developers to construct ML models with great scale, competence, and productivity all while filling model quality.
Old-style machine learning model development is resource-concentrated, needing important domain knowledge and time to yield and equate dozens of models. We’ll go faster the time it takes to acquire production-ready ML models with automated machine learning easily and with great productivity.
Azure Machine Learning deals the following two experiences to do work with automated Machine Learning. Find below sections to know feature availability in each experience.
- Azure Machine Learning Python SDK for code-experienced customers.
- Azure Machine Learning studio at https://ml.azure.com for limited/no-code experience customers.
When to use Automate Machine Learning
Put on automated Machine Learning when we want Azure Machine Learning to train and tune a model for us using the target metric we identify. Automated Machine Learning democratizes the machine learning model development process. It authorizes its users, no matter their data science know-how, to classify an end-to-end machine learning pipeline for any problem.
Machine Learning specialists and developers from corner to corner industries may use automated Machine Learning to:
- Apply ML solutions deprived of wide programming knowledge
- Time and resources saving
- Influence data science greatest practices
- Deliver agile problem-solving
This is a general and common machine learning task. It is a type of supervised learning in which models pick up using training data, and relate those learnings to new data. Azure Machine Learning deals featurizations exactly for these tasks, for example deep neural network text featurizers for classification.
The key objective of classification models is to predict which classes’ new data will drop into based on learnings from its training data. General classification instances are;
- Fraud detection
- Handwriting recognition
- Object detection.
More examples of classification and automated machine learning can be seen in the following Python notebooks.
Regression tasks are likewise a common supervised learning task similar to classification. Azure Machine Learning deals featurizations exactly for these tasks.
Regression models guess numerical output values built on independent predictors. The aim is to help found the association among those independent predictor variables by approximating how one variable influences the others in regression. For instance, automobile price founded on structures like, gas range, safety rating, etc.
Find more instances of regression and automated machine learning for predictions in the below Python notebooks.
Taking predictions is an important part of any business. It may be seen in revenue, inventory, sales, or customer demand. We can use automated Machine Learning to syndicate methods and approaches and get a suggested, high-quality time-series prediction. For more details with this how-to: automated machine learning for time series forecasting.
An automated time-series experiment is preserved as a multivariate regression problem. Historical time-series values are pivoted to develop extra dimensions for the regressor composed with other predictors. This method, different to classical time series methods, has a benefit of naturally joining many contextual variables and their affiliation to one another during training. Automated Machine Learning acquires a lone, but a lot within branched model for all items in the dataset and forecast horizons. Additional data is therefore obtainable to guess model parameters and generalization to hidden series develops possible.
The below are included in advanced forecasting configuration.
- holiday detection and featurization
- time-series and DNN learners (Auto-ARIMA, Prophet, ForecastTCN)
- many models support over grouping
- rolling-origin cross validation
- configurable lags
- rolling window aggregate features
Fine the examples of regression and automated machine learning for predictions in the following Python notebooks.
Computer vision tasks permit us to effortlessly make models trained on image data for scenarios similar to image classification and object detection.
We can perform following tasks with this capability.
- Without a glitch assimilate with the Azure Machine Learning data labeling capability
- Use labeled data for creating image models
- Enhance model performance by identifying the model algorithm and tuning the hyper parameters.
- Download or set up the resulting model as a web service in Azure Machine Learning.
- Operationalize at scale, leveraging Azure Machine Learning MLOps and ML Pipelines capabilities.
Natural language processing (NLP)
Natural language processing (NLP) jobs in automated Machine Learning allow us to simply make models trained on text data for text classification and named entity recognition scenarios. Authoring automated Machine learning trained NLP models is maintained via the Azure Machine Learning Python SDK.
We can perform the below tasks easily with the NLP capability.
- End-to-end deep neural network NLP training with the latest pre-trained BERT models
- Seamless integration with Azure Machine Learning data labeling
- Use labeled data for generating NLP models
- Multi-lingual support with 104 languages
- Distributed training with Horovod
How automated Machine Learning works
We can plan and run our automated ML training experiments with following steps by using Azure Machine Learning.
- Classify the ML problem to be solved: classification, forecasting, regression or computer vision.
- Select whether we want to use the Python SDK or the studio web experience: Pick up about the equivalence concerning the Python SDK and studio web experience.
- For limited or no code experience, try the Azure Machine Learning studio web experience at https://ml.azure.com
- For Python developers, check out the Azure Machine Learning Python SDK
- Identify the source and format of the labeled training data such as Numpy arrays or Pandas dataframe
- Configure the compute target for model training, for example the local computer, Azure Machine Learning Computes, remote VMs, or Azure Databricks.
- Configure the automated machine learning parameters that control how many iterations over different models, hyperparameter settings, advanced preprocessing or featurization, and what metrics to search when defining the greatest model.
- Submit the training run.
- Review the results