Multidimensional hierarchy level recommendation for forecasting models
Provided is a system and method for identifying and recommending the best hierarchical levels for training a predictive model such as a time-series forecasting model. In one example, the method may include receiving an identification of a measure of multidimensional data, generating a plurality of training data sets that comprise different combinations of hierarchical dimension granularities of aggregation, training a plurality of instances of a machine learning model based on the plurality of training data sets, respectively, and determining and outputting predictive accuracy values of the plurality of instances of the trained machine learning model.
1. A computing system comprising:
a processor configured to:
receive an identification of a measure of multidimensional data;
extract metadata from the multidimensional data, the extracted metadata identifying a dimension of the measure and a plurality of different granularities applicable to the dimension;
generate, based on the extracted metadata, a plurality of multidimensional queries that comprise a plurality of different granularities of aggregation for the measure identified in the extracted metadata;
query the multidimensional data via the plurality of multidimensional queries to generate a plurality of training data sets;
train a plurality of different instances of a machine learning model via execution of a machine learning algorithm on the plurality of training data sets corresponding to the plurality of different granularities of aggregation, respectively;
determine predictive accuracy values of the plurality of different instances of the machine learning model corresponding to the plurality of different granularities of aggregation;
generate a ranked ordering of the plurality of different instances of the machine learning model based on the predictive accuracy values; and
display, in accordance with the ranked ordering, the predictive accuracy values via a user interface.
2. The computing system of claim 1 , wherein the multidimensional data comprises a time dimension.
3. The computing system of claim 1 , wherein the processor is further configured to identify restrictions on feature generation from the metadata, and generate the plurality of training data sets based on the identified restrictions.
4. The computing system of claim 1 , wherein the processor is configured to generate at least two training data sets from a same cube of the multidimensional data using a different aggregation granularity for time, respectively.
5. The computing system of claim 1 , wherein the processor is configured to generate at least two training data sets from a same cube of the multidimensional data using a different aggregation granularity for the measure, respectively.
6. The computing system of claim 1 , wherein the processor is configured to, for a training data set, execute the machine learning algorithm on the respective training data set to generate a corresponding instance among the plurality of different instances of the machine learning model.
7. The computing system of claim 1 , wherein the processor is configured to determine a predictive accuracy of an instance of the machine learning model based on a comparison of a forecasted output of the instance of the machine learning model with an actual output.
8. The computing system of claim 1 , wherein the processor is further configured to arrange identifiers of the plurality of different instances of the machine learning model from most accurate to least accurate based on the predictive accuracy values, and display the arranged identifiers via the user interface.
9. A method comprising:
receiving an identification of a measure of multidimensional data;
extracting metadata from the multidimensional data, the extracted metadata identifying a dimension of the measure and a plurality of different granularities applicable to the dimension;
generating, based on the extracted metadata, a plurality of multidimensional data sets that comprise a plurality of different granularities of aggregation for the measure identified in the extracted metadata;
querying the multidimensional data via the plurality of multidimensional queries to generate a plurality of training data sets;
training a plurality of different instances of a machine learning model via execution of a machine learning algorithm on the plurality of training data sets corresponding to the plurality of different granularities of aggregation, respectively;
determining predictive accuracy values of the plurality of different instances of the machine learning model corresponding to the plurality of different granularities of aggregation;
generating a ranked ordering of the plurality of different instances of the machine learning model based on the predictive accuracy values; and
displaying, in accordance with the ranked ordering, the predictive accuracy values via a user interface.
10. The method of claim 9 , wherein the multidimensional data comprises a time dimension.
11. The method of claim 9 , wherein the method further comprises identifying restrictions on feature generation from the metadata, and generating the plurality of training data sets based on the identified restrictions.
12. The method of claim 9 , wherein the generating comprises generating at least two training data sets from a same cube of the multidimensional data using a different aggregation granularity for time, respectively.
13. The method of claim 9 , wherein the generating comprises generating at least two training data sets from a same cube of the multidimensional data using a different aggregation granularity for the measure, respectively.
14. The method of claim 9 , wherein the training comprises, for a training data set, executing the machine learning algorithm on the respective training data set to generate a corresponding instance among the plurality of different instances of the machine learning model.
15. The method of claim 9 , wherein the determining comprises determining a predictive accuracy of an instance of the machine learning model based on a comparison of a forecasted output of the instance of the machine learning model with actual output.
16. The method of claim 9 , wherein the outputting comprises arranging identifiers of the plurality of different instances of the machine learning model from most accurate to least accurate based on the predictive accuracy values, and displaying the arranged identifiers via the user interface.
17. A non-transitory computer-readable medium comprising instructions which when read by a processor cause a computer to perform a method comprising:
receiving an identification of a measure of multidimensional data;
extracting metadata from the multidimensional data, the extracted metadata identifying a dimension of the measure and a plurality of different granularities applicable to the dimension;
generating, based on the extracted metadata, a plurality of multidimensional queries that comprise a plurality of different granularities of aggregation for the measure identified in the extracted metadata;
querying the multidimensional data via the plurality of multidimensional queries to generate a plurality of training data sets;
training a plurality of different instances of a machine learning model via execution of a machine learning algorithm on the plurality of training data sets corresponding to the plurality of different granularities of aggregation, respectively;
determining predictive accuracy values of the plurality of different instances of the machine learning model corresponding to the plurality of different granularities of aggregation;
generating a ranked ordering of the plurality of different instances of the machine learning model based on the predictive accuracy values; and
displaying, in accordance with the ranked ordering, the predictive accuracy values via a user interface.
18. The non-transitory computer-readable medium of claim 17 , wherein the multidimensional data comprises a time dimension.
19. The non-transitory computer-readable medium of claim 17 ,
wherein the method further comprises identifying restrictions on feature generation from the metadata, and generating the plurality of training data sets based on the identified restrictions.
20. The non-transitory computer-readable medium of claim 17 , wherein the generating comprises generating at least two training data sets from a same cube of the multidimensional data using a different aggregation granularity for time, respectively.