Neuron

Spark takes Excel files and converts them to code, accessed via application programming interface (API) services. Neuron is the name of our next-generation "Excel-to-code" service type!

Neuron significantly enhances our past implementations of "Excel-to-code" by converting Excel files into fast and efficient C++ code which is packaged into WebAssembly (Wasm) modules. These Wasm modules enable many use cases including:

• Single calculations in the Execute API.

• Scaled execution via the Testing Center or Batch APIs.

• Bundling code into Hybrid Runner.

• Built into an application with SDKs.

Best practices with Neuron

For an optimal Neuron compilation of an Excel file, it is recommended to follow the guidance in:

• How to: Prepare an Excel file for Spark

• Excel tips

Neuron versions

Neuron version for a service version

With each Spark release, a new version of Neuron is deployed with enhancements and fixes. To view the Neuron version used for a particular service version:

• From API Tester, submit an API request with Neuron and Review API responses. The Transaction details will state the compiler_version used to convert Excel to code.

• The same details can be seen in the Execute API Execute API.

Managing Neuron versions

• For information about Neuron version management see: Choose the Neuron compiler for newly uploaded services, Choose the Neuron compiler version for additional service versions, and Neuron release history.

• For information about recompiling your service with a different version of Neuron see Recompile service.

Functional limitations

Known issues

If you encounter an issue with an Excel file that does not convert to a Spark service, please contact Support.

• Forms and Controls, Visual Basic for Applications (VBA), Office Scripts, and Python in Excel are not supported.

• Excel maintains values at 15 significant digits, learn more in Numeric precision in Microsoft Excel.

• Cells containing external links will have their links broken and the value saved into the file at the point of upload will be used for that cell. It is recommended to break links in advance of uploading to Spark where possible.

• Not every formula will work correctly with array formulas.

• Although we do support complex LAMBDA() functions, not every function definition may be supported. This also applies to the associated helper functions.

• Pivot Tables, Power Pivot, Power Query, Data Models, Scenario Manager, Solver, Table slicers are not supported.

• Use 1904 date system is not supported.

• Lotus Compatibility Settings for Transition formula evaluation is not supported and may lead to erroneous results when used with Spark.

Other considerations

• There is support for dynamic array formulas and spilled arrays, however inputs and outputs must be set to a predefined size. Setting an input or output with a spill range operator # such as D3# is not supported.

• For the NOW(), RAND(), RANDARRAY(), RANDBETWEEN(), and TODAY() function(s), the calculation results may vary with each calculation. Dates and times may not be relatable to each tenant's deployment. NOW()and TODAY() can be overridden using the Execute API (v3) request_meta manual_now.

• When using d AVERAGEIFS, COUNTIFS, and SUMIFS functions, be sure the criteria_range and criteria use the same datatypes. i.e. If the criteria is a text value, then all the cells in the criteria_range should also be text. Otherwise, inconsistent and unexpected results may be returned.

• The INDIRECT() function limits calculation performance optimization opportunities. Consider rewriting this formula to use a combination of CHOOSE(), MATCH(), INDEX(), and/or OFFSET().

Compiler messages

Neuron generates a number of messages in the Review the Upload log. For more information about these messages, see Upload log messages.