dotnet / dotnet/machinelearning-modelbuilder
Suggestion: Shareable dataset schema
- Dominant language
- Dockerfile
- Stars
- 285
- Forks
- 66
- PR merge metrics
- No merged PRs in 30d
Description
**Is your feature request related to a problem? Please describe.**
Sometimes the InferColumns from ML.NET and Model Builder assume the data type differently. Also, in some cases I might want to change 0/1 values to Boolean manually, or vice versa. One model with Model Builder also has boolean ("true"/"false") as a String, while other detect it as Boolean. I dont think I changed it manually, but it is possible. Also could because of different Model Builder versions used to create it.
This causes minor issues because I am not able to use the consumer client code to run the models.
**Describe the solution you'd like**
A file describing the column types and purposes for both Model Builder and ML.NET library. .MBConfig file could be suitable for this purpose.
For model builder, user can copy the existing .mbconfig, but there maybe could be another way to "copy settings from..."
In my work, it does not have a big impact but requires small changes when running the models.
**Additional info**
Maybe there could also be a way in ML.NET to help auto-convert data types when enabled. For example {0,1} could be Boolean, String or Single and the conversion could be done.
It might also be advantegous that ML.NET class library and Model Builder would give same results when inferring field types.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.