validation
Hi,
What are types of validation in Auto Model (Bootstrapping Validation, Cross Validation, Split Validation, Wrapper Split Validation, Wrapper-X-Validation) ?
What are types of validation in Auto Model (Bootstrapping Validation, Cross Validation, Split Validation, Wrapper Split Validation, Wrapper-X-Validation) ?
Find more posts tagged with
Sort by:
1 - 5 of
51
As I can see the Auto Models are validated using multiple hold-out validations, which are then averaged using Performance Average (Robust). So this is not a "standard" RM process. When you generate an Auto Model, you can inspect each process individually by clicking into "Open Process".
Hello @vps
Its a multi hold out validation as informed by @jacobcybulski. The data is split into 60:40 (train:test) ratio. The 40 % of test data is again split into 7 subsets and then all the 7 subsets are tested and their performances are aggregated to get the final performance in auto model.
Its a multi hold out validation as informed by @jacobcybulski. The data is split into 60:40 (train:test) ratio. The 40 % of test data is again split into 7 subsets and then all the 7 subsets are tested and their performances are aggregated to get the final performance in auto model.
Thanks!
Is there any documentation regarding the validation in Auto Model?
HI @vps,
Yes, there is a short description of how is validated the model in the final screen of Auto-Model (the results screen):
just click on the "information (i)" button :
Regards,
Lionel
Yes, there is a short description of how is validated the model in the final screen of Auto-Model (the results screen):
just click on the "information (i)" button :
Regards,
Lionel
Thanks a lot, Lionel. That seems to completely fix the issue.
Sort by:
1 - 3 of
31
As I can see the Auto Models are validated using multiple hold-out validations, which are then averaged using Performance Average (Robust). So this is not a "standard" RM process. When you generate an Auto Model, you can inspect each process individually by clicking into "Open Process".
Hello @vps
Its a multi hold out validation as informed by @jacobcybulski. The data is split into 60:40 (train:test) ratio. The 40 % of test data is again split into 7 subsets and then all the 7 subsets are tested and their performances are aggregated to get the final performance in auto model.
Its a multi hold out validation as informed by @jacobcybulski. The data is split into 60:40 (train:test) ratio. The 40 % of test data is again split into 7 subsets and then all the 7 subsets are tested and their performances are aggregated to get the final performance in auto model.
HI @vps,
Yes, there is a short description of how is validated the model in the final screen of Auto-Model (the results screen):
just click on the "information (i)" button :
Regards,
Lionel
Yes, there is a short description of how is validated the model in the final screen of Auto-Model (the results screen):
just click on the "information (i)" button :
Regards,
Lionel