Chronos is a fully-featured time series forecasting and analysis platform. We offer solutions that take users from data creation, data exploration, all the way to using advanced forecasting algorithms involving deep learning and machine learning.
Chronos can be broken down into three main sections:
These sections are helpful for gaining an understanding of how to break down the process of working with time series from end-to-end, but it’s not exactly how the platform is built. Instead of there being a strict linear path to follow from start to finish, we make it so that users have the ability to move freely in the path, starting with uploading their data to making a forecast. We’ll get more detail as to what this means, but the essence is that a time series is just an ordered series of numbers over time. How and where this data is used is left to the user - this allows for more creativity and human intuition compared to a previously linear and strictly statistical process. At the same time, this documentation should leave users with enough guidance that they feel comfortable taking the process on themselves.
Chronos is made up of a number of apps that each provide the user with an interface to accomplish some part of the time series forecasting and analysis process. The order we introduce these in is only one way the user can use the apps. We’ve built the platform in a flexible manner so that the user maintains full control and that no assumptions are made about how the platform is used and what kind of data is being worked with.
We currently support
.xlsx file uploads and the primary way for users to bring their own data onto the platform. This is usually the first stop a user makes before starting to use the rest of the platform.
Formatting your data before bringing it on to the platform is essential. We've done as much as we can to accept many different formats and variations of data but this step is none-the-less important. Tools like Excel or Google Sheets are good places to work with your time series data to ensure the formatting is correct. Often there are utilities to transform data all at once so the process should be quick and easy.
Data should look like the tables below: column name "date" must be present, however the "values" column name can be called anything
In some cases you'll want to bring in more than one series at a time, for example with Batch forecasting. To specify additional series, add the values as a new column while maintaining a single date column. If the series' date ranges do not line up, that is okay. Include all the dates need and cells where a series doesn't have values can be left blank.
An example of this could look like the following:
The table above also shows multiple series with differing date ranges. series_one spans from
2022-01-04 and series_two spans from
2022-01-06. Chronos will handle these cases without a problem as long as the date frequency of all the series being uploaded is the same. For example, mixing monthly data and daily data will not work.
Precision forecasting is all about generating predictions for a single time series. Most commonly, these predictions will be into the future so that users of the forecast can make more informed decisions.
The process for precision forecasting is simple. Choose some data, choose how far into the future you’d like to forecast, choose an algorithm, and let Chronos do the rest. Check out the quick start documentation for an interactive guide through the basics of generating a forecast on Chronos.
Once you submit a forecast, Chronos will do a number of things before returning the forecast back to you. First, it takes your data and ensures that it’s correctly formatted. If not, it will make a number of attempts to correct any issues before moving on to the next step. Once we can ensure that the data can be forecasted, the user selected algorithm takes over. Each algorithm has its strengths and weaknesses. We do our best to share what we’ve learned in this documentation, but in the end we leave a lot of the decisions up to the user. Through thorough experimentation, Chronos users will build up an intuition as to what works and what doesn’t. Much of these decisions should be based on the data being forecasted and likely no one will understand that data better than the user.
Scenario-based forecasting combines the power of precision forecasting with multiple scenarios to provide a tool that accels at helping users understand the most likely outcome when there is a great deal of uncertainty about the future.
Users upload their target series, but also include one or more scenarios they’ve created externally or with the Scenario Creation tool. Scenarios are a collection of one or more time series that have values extending into the target series forecast horizon. Chronos will individually forecast the target with each of the scenarios to get a range of possible forecast outcomes.
Given these outcomes and the scenarios that informed them, users get a better sense of how their scenarios move the target and in turn it provides more information when making decisions based off of the target forecast.
Scenario Creation is a fundamental tool to scenario-based forecasting. The tool allows users to build scenarios that capture what they think will happen in the future. Building off of their own or GlobalPulse data, we have historical observations that we then must continue into the future. Since the future is unknown, a user’s subject-matter expertise and intuition are an essential part to generate these future values.
Often, a general sense of increasing or decreasing is enough for the scenario to be effective at driving the direction of a forecast that uses that scenario.
With the Scenario Creation tool, users can quickly generate these scenarios to be used in their forecasts.
Data Exploration is a tool that enables our users to gain a deeper understanding of their own data. It provides a number of alternative visualizations and statistical/model-based transformations of the data that can uncover insights without having even forecasted the data yet.
While most of Chronos is focused on forecasting and looking into the future, we also recognize the importance of understanding the past. Data exploration shows users the underlying patterns and features of your data that eventually will be used to forecast into the future. But by seeing these features yourself, you can gain a sense of how forecastable your data might be.
There are a number of insights that come from data exploration that can be of service to the user. Depending on the forecasting algorithm, you may or may not get an explanation of how the forecast results are generated. With data exploration, you can see the patterns for yourself.
GlobalPulse is a separate service in the Nousot Ecosystem that has been integrated into Chronos. The goal of GlobalPulse is to provide users with a hand picked collection of dataset that will aid their data work. In Chronos, we get a subset of GlobalPulse limited to time series data. Using these datasets, users can choose to either forecast them directly or to use them as a base to build a scenario to be used in scenario-based forecasting.
In coming development, there will also be pre-forecasted datasets. Our team of subject-matter experts come together to share a forecast that the community can use as a scenario in their own forecasts. Over time, trust can built up in the Chronos team’s forecasts, driving forecasts and in the end giving our users more trust in their own forecasts and the decisions they make based off of them.
Modern data science and machine learning often gloss over forecasting. Despite this, forecasting remains a relevant problem to nearly every industry and business.
Businesses all over the world track numerous metrics and data points that change over time. This is the essence of what a time series is. Given a time series, algorithms attempt to model the behavior of that data. Optionally, additional data may be used that can give more insight into the patterns in that data. With enough of a history to train on, a model can generalize the patterns within a time series well enough that it can make predictions about the future.
These future predictions will not always be accurate, but not all is lost. The benefit of time series is that as time passes and new data is collected, eventually we will be able to see how predictions compare to the true values. This sets up the idea feedback loop that allows us to continually improve our predictions until enough trust has been built up in the predictions.
Given this trust, leaders can make business decisions based on the predictions. This insight creates businesses that are anticipatory, rather than reactionary. Staying ahead of the curve is how businesses can maximize their impact in all areas.