The following steps and items are part of the onboarding process:

1. Create a CME Group ID following these instructions. 

2. Log-in to DataMine. 

3. Navigate to the DataMine Machine Learning Service here.

4.  

   1. Here you will be able to subscribe.

5. Go to your cart and checkout. 

6. You will now be redirected to the licensing workflow. To proceed you will need to provide the below details:

7.  

   1. Name of Individual or Company

   2. Individual/Company primary address 

   3. Individual/Company billing address 

   4. Contact information for at least two people

   5. Usage details: Internal (using the data inside your company only), Redistribution (sending the data outside your company), Derived (unique data created by making material calculations and / or changes to the licensed data, so that the licensed data cannot be substituted by, reverse engineered or otherwise identified from the unique data).

   6.  

      1. If Redistribution or Derived usage is required, additional licensing steps will be necessary.

8. Upon completion of step five, you will be required to sign the DataMine agreement via DocuSign or click-through.

9.  

   1. Please note, we do not allow this agreement to be modified. 

10. After signature, the order can take up to 48 hours to be approved. 

11. Once the order is approved, you can access the service here. 

Who can use DataMine Machine Learning Service?

DataMine Machine Learning Service is designed to make machine learning accessible to non-technical users. With Machine Learning Service you don’t have to be a Data Scientist to utilize machine learning.

Is DataMine Machine Learning Service also helpful for data scientists and other more advanced users?

Yes, DataMine Machine Learning Service streamlines and automates many of the steps in machine learning workflows, freeing up more advanced users to work on higher-value tasks such as signals processing and strategy implementation.

How far into the future can DataMine Machine Learning Service predict?

Machine Learning Service can build models for any timeframe. Different timeframes generally benefit from different data inputs.

How often are the models updated?

Models are updated every day. New data is ingested overnight, and the models update first thing in the morning to be available on the US market open.  

How many models can I build using the DataMine Machine Learning Service?

There is no limit to the number of models a user can build.

How is model performance measured?

The DataMine Machine Learning Service is a tool. All models are measured via performance metrics and backtesting in order for users to make their own evaluation of the model performance and establish confidence in the tool.

 What is the Markets Page? 

The Markets Page is a high-level view of all available forecasts across categories and sectors. The primary goal of the Markets Page is to give users a directional indication of different markets. Additionally, users can toggle on or off labels for Signals and Accuracy, to see the respective signal strength and accuracy of forecasts at different time intervals.

What is the Forecast Page? 

The Forecast Page is a medium-level view of forecasts for specific target variables. The purpose of this page is to provide users with a more detailed illustration of the forecast shape (Prediction Curve), behavior and performance at different time intervals, as well as a historical record of the forecast backtesting over time.

What is the Strategy Simulator Page?

The Strategy Simulator page is a tool that uses signal processing to turn model signals of various timeframes into actionable insights. It does this by combining signals from multiple models using different parameters and filters, so only the best model signals are used. Users are provided with default strategy parameters, but can further optimize strategies to their own individual goals, strategy preferences, and risk profile. Examples of strategy preferences:

• Select which model signals are used in the strategy. 

• Optimize for different performance metrics such as NetPosition (Equity Curve), ROI, Sharpe Ratio, Information Ratio, Max Drawdown, etc. 

• Filter out model signals below certain performance thresholds. 

• Trade Long/Short, Long-Cash, or Short-Cash.

• Trade only signals above or below certain thresholds. 

What is the Data Explorer Page?

The Data Explorer is a tool designed to aid users in exploring and navigating CME Group data available for modeling. The dashboard provides a classification Search function as well as helpful visual elements to help the user find what they are looking for, and see the structure and categorization of the available data.

What is Autoseries?

Autoseries is a model building method that builds many machine learning models in a challenger champion framework. The challenger champion framework helps the tool adapt to changing marketing conditions, where different models compete for the top spot. Every month the models are reevaluated, and a new champion (better model) is selected. The outputs from the champion models are stitched together to give the impression of one continuous model. The reason why it is beneficial to build more models is because it allows the models to explore a larger algorithm parameter space and converge on a better output.

How is DataMine Machine Learning Service hosted?

The DataMine Machine Learning Service is hosted as a managed cloud solution.

How much does DataMine Machine Learning Service cost?

The DataMine Machine Learning Service is charged on a per-model basis. Each model is $20 to create and every day it runs, an additional $2/day fee will be applied. For example, if you created two models and each ran for 30  days you would pay: ($20/model *2) + ($2/day * 30 days *2 models) = $160

Can I get trial access to the DataMine Machine Learning Service?

Yes, there is a one-week trial available for all new users. You must opt-out of the trial to avoid being charged. Contact a sales agent at cmedatasales for more details.

How many users can DataMine Machine Learning Service support?

The DataMine Machine Learning Service can support any number of users. Licenses are sold on an individual basis.

How many machine learning algorithms does DataMine Machine Learning Service use?

DataMine Machine Learning Service currently offers a library of algorithms as options when building a machine learning model:

Is the input data included in the license cost?

Yes, input data is included in the license cost. Users are only charged for building models. 

What data is available through The Machine Learning Service?

How are the CME Group Rolling Futures Indices calculated?

The CME Group Rolling Futures Indices are designed to provide market participants with an indicator for investment performance in a single CME Group Market.

CME Group Rolling Futures Indices are calculated on 28 futures products across five different asset classes. The CME Group Rolling Futures Indices will be calculated for each Business Day, in accordance with the CME Group Globex Trading Schedule and follow a next-month rolling schedule.

To maintain a continuity or pricing and account for contract expiry and movements in liquidity, a roll period is required. The roll period is designed to gradually transfer the weighting of the inputs from the Near Futures Contract to the Second Futures Contract prior to expiry. The weighting of the Near Futures Contract is 100% up until seven days to expiry. After this point, 20% weighting is transferred to the Second Futures Contract on a daily basis. When the Near Futures Contract is 2 days to expiry, it will have a weight of 0% and the Second Futures Contract will have a weight of 100%.

CME Group Rolling Futures Indices are calculated for a list of CME Group products.

What frequency of data can/do the models use? 

Machine Learning Service is capable of using data of different frequencies. While the models themselves update daily, the input data frequency ranges from daily, to weekly, monthly, quarterly, or annual data resolution. 

Can I use my own data?

Currently the solution does not support users ingesting their own data.

Can I build my own models?

Yes, users can build and maintain their own machine learning models. 

How are models trained?

Models are trained on 5 years of historical data preceding the model Start Date (assuming the data is available). For example, if a model is built from January 1, 2022, the model will be trained on historical data from January 1, 2017. If historical data is not available going back that far, the model will train on what it is provided. 

How are the models back-tested?

Model metrics are displayed in the dashboard to facilitate straightforward analysis and evaluation of model performance. 

How do I assess model performance?

The Machine Learning Service provides a list of performance metrics that are updated daily as a part of the model building/update process.

Are my models saved?

With Machine Learning Service, all user-created models are stored on the cloud. Models are backed up to hard storage periodically in case of system failure.