How long does it take to build an ML model?

November 16, 2021 · 3 minute read

John Patrick Hinek

Growth

TLDR

There is no simple equation that can solve for the time it will take to build an ML model. The answer will depend on what data and model is being used, and intended accuracy.

What goes into creating a machine learning model

Machine learning (ML) has become widely used for its efficiency in assisting with data analysis and bettering workflows. As ML becomes integrated within different stages of business, understanding the time it takes to deploy ML models becomes essential for project management. Unfortunately, there isn’t an equation that can dictate the time it takes to build and deploy these models. As ML is still in its infancy, deploying models is still not something that happens very quickly. According to

Algorithmia’s “2020 State of Enterprise Machine Learning”

, 50% of respondents said it took 8–90 days to deploy one model, with only 14% saying they could deploy in less than a week.

According to Algorithmia’s findings, the maturity of a company’s ML program was the biggest factor in determining how quickly they could deploy their models. Having more seniority in the space allows for companies to build up their teams, data, and systems to be able to deploy models quicker.

Algorithmia’s 2020 graph on the relation between model development time and company size.

The quicker that models are deployed, the better. ML used to make more informed business decisions falls short if it can’t deliver fast and accurate results from the data. It is often not cost-effective to use ML if companies aren’t able to deploy models inline with emerging business operations.

Along with seniority in the ML space, there may be other contributing factors as to why models aren’t being deployed faster. As models are run on data, having access to the right kind of data is a key factor when deploying models. Models run best on structured data, which takes time to clean through. First time users of ML may find it difficult to gather or retrieve enough relevant data, if not historically stored correctly. Having a large enough dataset is essential to creating an effective model. However, more data means more computing power and time required to execute the project.

The runtime to generate models can be very time-consuming depending on the amount of data it’s given and the desired accuracy. Accuracy is measured by the number of correct predictions divided by the number of total predictions made. Feeding in a large amount of data and getting enough correct predictions to be statistically significant can often take a great deal of time.

While not all models will require extremely high accuracy, models built for detailed and important use cases, such as in health care, must be matched with a very high level of accuracy.

With more innovation within the AI space, the time it takes to create a model shows signs of going down.

Amazon SageMaker

is making steps to close this gap helping users build, train, and deploy models. Offering a wide range of frameworks and algorithms, SageMaker will train the model and help to deploy it to production. Tools such as SageMaker can cut down the steps done by the developer to save time when building a model.

When generating your own ML model, things to consider when managing time are your developers familiarity with creating models, the type of data you’re using and its size, and the accuracy needed to run the project. When done correctly, ML can produce great results and generate more informed business decisions. With an increase in funding and innovation showing no sign of slowing down, the average time it takes to deploy an ML model is likely to keep going down.

Hang out with us

Join our community and chat about startups, AI/ML, and product development.

Like what you see? Join the team.

Mage is making AI and ML accessible to product developers. Join us and build beautiful and intuitive devtools.

Want to give us feedback or ask questions?

Please chat with us live by joining our Discord channel or send us an email.