Google brings machine learning to online spreadsheets with Simple ML for Sheets

Spreadsheets Used by organizations of all sizes for a variety of basic and complex tasks.

While simple calculations and graphs have long been part of the spreadsheet experience, machine learning (machine learning) No. ML is often considered too complex to use, whereas the use of spreadsheets is designed to be accessible to any type of user. Google Now is trying to change the paradigm with its Google Sheets online spreadsheet program.

Today Google announced the beta Simple ML for Sheets plugin. Google Sheets has an extensible architecture that enables users to benefit from add-ons that extend the default functionality available in the application.In this case, Google Sheets benefits from ML technology first developed by Google in open source tensor flow project. With Simple ML for Sheets, users don’t need to use a specific TensorFlow serving, as Google has developed the service to be as accessible as possible.

“Everything runs entirely on the user’s browser,” Google AI development advocate Luiz Gustavo Martins told VentureBeat. “Your data never leaves Google Sheets, and models are saved to your Google Drive so you can use them again later.”

Gosh, what can Google’s Simple ML do with my spreadsheet?

So what can Simple ML for Sheets do? Two of the beginner tasks in the beta that Google highlighted include the ability to predict missing values ​​or spot outliers. These two beginner tasks make it easy for anyone to test the ML waters and explore how ML can benefit their business, Martins said.

Martins noted that in addition to the starter tasks, the plugin also supports several other common ML tasks, such as training and evaluating models, generating predictions, and interpreting models and their predictions. Also, since Simple ML can export models to TensorFlow, people with programming experience can use Simple ML models with their existing ML infrastructure.

Overcome the challenges of ML complexity with Simple ML for Sheets

Google Sheets users can benefit from ML without Simple ML, but it might not be easy for a layman.

“We think knowledge and lack of guidance are the main factors that make it easy for non-ML practitioners to use ML,” Google software engineer Mathieu Guillame-Bert told VentureBeat. “Using classic ML tools, such as TensorFlow in Python, is like being on a blank page.”

Using classic ML tools requires users to understand programming, ML problem framing, model building and model evaluation, Guillame-Bert said. He noted that this knowledge is often acquired through classrooms or long periods of self-study.

In contrast, Guillame-Bert said Simple ML is like an interactive questionnaire. It guides the user and assumes only basic knowledge about spreadsheets.

Powering Simple ML with Decision Forests

Martins explained that behind the scenes, the Simple ML plugin uses Yggdrasil Decision Forest Library. This is the same library TensorFlow Decision Forest.

“For this reason, once trained in the plugin, advanced users can export models to any TensorFlow Serving hosting service, such as TensorFlow Serving on Google Cloud,” Martins said.

TensorFlow Decision Forests (TF-DF) is a library of algorithms for training new models, Guillame-Bert explained. In other words, users provide examples to TF-DF, and they receive a model in return. He points out that TF-DF does not have pretrained models; however, since TF-DF is integrated in the TensorFlow ecosystem, advanced users can use decision forests with pretrained models.

according to published researchthe technique behind TF-DF is based on the concepts of random forests and gradient boosted trees, which are well suited for training models on tabular datasets such as spreadsheets.

Going forward, Guillame-Bert said Google will work to further improve the availability of add-ons. Google also plans to add new features to Simple ML for Sheets that don’t require users to have any knowledge of ML.

“During internal testing, we identified several high-demand tasks that we thought would be popular with users,” Guillame-Bert said. “We hope to gain feedback from this public release to prioritize and design these missions.”