FAQ and Common Issues#
In case of issues with Simple ML for Sheets, please check out the known issues section of this FAQ.
If your issue is not listed, the best way to find help is through the Simple ML for Sheets User group
If you found a bug, please report it. In last resort, contact us directly.
About Simple ML for Sheets#
Q: What is Simple ML for Sheets?#
Simple ML for Sheets is an Addon for Google Sheets that allows to use machine learning without being an ML expert, without coding, and without sharing data with third parties.
Q: Who developed Simple ML?#
Simple ML is developed by the TensorFlow Decision Forests team in Google Zurich.
Q: Where does my data go when using Simple ML?#
Data and models are in your Google Sheets and in your Google Drive. All the operations (including model training operations) are executed on your computer in your web browser. That’s why your data is not exported to any third-party server.
Q: Where are my ML models stored?#
Models are saved on your Google Drive in the folder simple_ml_for_sheets.
Q: What technology is used by the Simple ML?#
Simple ML uses decision forest models (e.g., Random Forests, Gradient Boosted Trees) as those models are particularly well suited for tabular datasets. For more information about Decision Forests, we recommend our Decision Forests class on MLCC.
Under the hood, Simple ML is powered by Yggdrasil Decision Forests, a fast, flexible and powerful library for state-of-the-art Decision Forest algorithms. To learn more about the library, check out the YDF documentation.
The forecasting is done under the hood by ensembling two ARIMA and two ETS models.
Q: Can I export my model in Colab and use it in Python?#
Yes. After you trained a model, select the “Export model” task, select the “Colab” option, and click on “Export”. You will see a snippet of Python code to copy/paste into Colab to run your model.
The resulting model is both a TensorFlow Decision Forests modeland an Yggdrasil Decision Forests model. It is compatible with TensorFlow Serving and the Yggdrasil Serving APIs.
Q: I have a problem and the documentation does not help me. What should I do?#
The best way to find help is through the Simple ML for Sheets User group
If you found a bug, please report it. In last resort, contact us directly.
Q: I found a bug. Where do I report it?#
To report a bug, you can use the report form
Q: Can I participate in this project?#
Yes :). The Simple ML for Google Sheets Addon is a new project, and we are looking for new ideas and new people.
If you have an algorithm that you want to see added to the addon (maybe this is an algorithm developed by your team), contact us.
Q: How is Simple ML able to predict the missing values?#
Simple ML trains a model on the non-missing values, and use this model to predict the missing ones.
Q: How is Simple ML able to detect abnormal values?#
Simple ML trains multiple models to predict the existing values using cross-validation. The predictions of those models are then compared to the actual values. If the predicted and actual values differ, the existing value is tagged as abnormal.
The abnormality is not a yes/no question. Instead, the abnormality is as a probability between 0% (normal) and 100% (abnormal).
Q: How do I interpret the abnormality score?#
The computation of the abnormal score depends on the task of the model (e.g. classification, regression). However, in all cases, it is between 0 and 1. With 0=completely normal, and 1=completely abnormal.
Check the Spot abnormal values documentation for more details.
Q: How is Simple ML able to forecast future values?#
Simple ML trains a forecasting model on the past values and identifies seasonality patterns and trends in the data. It also uses calendar data (e.g. holidays) to improve its forecast. Simple ML then applies this model to the new data.
Q: Can I export my forecasting model?#
Exporting forecasting models is not supported at this time.
Q: What is the relation between Simple ML and TensorFlow?#
Simple ML is based on the Yggdrasil Decision Forests library that is also used to power TensorFlow Decision Forests. That’s why the models trained on Simple ML are compatible with TensorFlow and Tensorflow Decision Forests.
Q: What permissions does Simple ML require and why?#
Simple ML requires the following permissions:
See, edit, create and delete all of your Google Drive files: Simple ML exports models to the user’s Google Drive.
See, edit, create, and delete all your Google Sheets spreadsheets: Simple ML can add new columns to your sheets (for example, to report predictions).
Display and run third-party web content prompts and sidebars inside Google applications: Simple ML is a sidebar within Google Sheets.
Q: If my data has some filter applied to it, which data does Simple ML use for training? The filtered one or the full data?#
Simple ML will use all the data. If you only want to use the filtered view, copy it to another tab and run the task on this new tab.
Known issues#
Every task fails with “The task failed. Fix the problem and run it again.”. The Source columns continuously shows a loading bar.#
This issue can happen if logged in with two Google accounts into Chrome and / or Google Sheets. We are actively working on a resolution. As a workaround, please try logging in with just one account when using Simple ML for Sheets.
The column I want to autocomplete is composed of 0’s and 1’s. When predicting missing values, the results are numbers between 0 and 1.#
By default, the type of problem (e.g., classification, regression) is detected automatically from the label. If your labels look like numbers (e.g., 0 1) it will be considered as regression task. You could change the label to “label_0” and “label_1” instead. This will change the task to classification and autocomplete with label_0 and label_1. To do this change, you can apply a formula like “=if(J1=0, “label_0”, “label_1”)”
Forecasting fails even though my dates are always the same day of the month, e.g. the 28th of each month#
Since the lengths of the months are variable, Simple ML for Sheets can sometimes struggle to detect if intervals really are monthly. If the event to forcast is monthly, prefer to set the first day of the month in the date column.