Automatic generation of labeled data in IoT systems
A labeled data generation service provides an Internet-of-Things (IoT) system with a capability whereby users may configure how the system gathers, processes, and generates labeled data instances by: collecting and processing the data into a format required by supervised learning algorithms; generating expected outputs from data available in the IoT system; supporting the linking of collected inputs with generated expected outputs; forming labeled data instances; cleaning the labeled data set appropriately; sending the labeled data set to target nodes; and/or communicating with target nodes regarding improving the data processing and labeling processes, as required.
1. An apparatus comprising one or more processors and memory storing computer-executable instructions which, when executed by the one or more processors of the apparatus, cause the apparatus to perform operations comprising:
maintaining a configuration, the configuration comprising design information for a labeled data set, the labeled data set comprising a plurality of labeled data instances, wherein each labeled data instance comprises a plurality of data values, the data values comprising one or more data inputs and one or more expected data outputs associated with the one or more data inputs;
acquiring a plurality of raw data inputs from data sources;
processing, according to the configuration, the raw data inputs to create processed data inputs, wherein the processing of the raw data inputs comprises pre-processing based on first parameters indicated in the configuration and data transformation based on second parameters indicated in the configuration;
generating, according to the configuration, labeled data instances, wherein a labeled data instance comprises one or more processed data inputs and one or more expected data output values;
storing the labeled instances in a labeled data set; and
sending the labeled data set to a repository.
2. The apparatus of claim 1 , wherein, for one or more raw data inputs, the processing of the raw data inputs comprises converting or scaling each of the raw data inputs to create a processed data input value for each raw data input.
3. The apparatus of claim 1 , wherein, for one or more raw data inputs, the processing of the raw data inputs comprises scaling a processed data input value for each of the raw data inputs in accordance with one or more statistical observations of the plurality of raw data inputs.
4. The apparatus of claim 1 , wherein, for one or more sets of raw data inputs, the processing of the raw data inputs comprises deriving a processed data input value for each plurality of raw data inputs in accordance with one or more statistical observations of the plurality of raw data inputs.
5. The apparatus of claim 1 , wherein the labeled data instances are generated with data cleaning based on one or more cleaning rules indicated by the configuration.
6. The apparatus of claim 5 , wherein the data cleaning comprises one or more of:
identifying duplicate labeled data instances in the labeled data set;
removing the identified duplicate labeled data instances from the labeled data set;
verifying data is valid from the labeled data set;
monitoring for mandatory data in the labeled data set;
detecting conflicts with data instances in the labeled data set; and
informing the repository of the identified duplicate labeled data instances.
7. The apparatus of claim 1 , wherein:
the configuration comprises an output time requirement parameter; and
the operations further comprise acquiring an expected data output in accordance with the output time requirement parameter.
8. The apparatus of claim 1 , wherein the pre-processing comprises one or more of: measurement unit conversion, data type conversion, or data aggregation.
9. The apparatus of claim 8 , wherein the data aggregation comprises one or more of: a sum, an average, a minimum, a maximum, or a count.
10. The apparatus of claim 1 , wherein the data transformation comprises one or more of: normalization, standardization, or binning.
11. A method comprising:
maintaining a configuration, the configuration comprising design information for a labeled data set, the labeled data set comprising a plurality of labeled data instances, wherein each labeled data instance comprises a plurality of data values, the data values comprising one or more data inputs and one or more expected data outputs associated with the one or more data inputs;
acquiring a plurality of raw data inputs from data sources;
processing, according to the configuration, the raw data inputs to create processed data inputs, wherein the processing of the raw data inputs comprises pre-processing based on first parameters indicated in the configuration and data transformation based on second parameters indicated in the configuration;
generating, according to the configuration, labeled data instances, wherein a labeled data instance comprises one or more processed data inputs and one or more expected data output values;
storing the labeled data instances in a labeled data set; and
sending the labeled data set to a repository.
12. The method of claim 11 , wherein, for one or more raw data inputs, the processing of the raw data inputs comprises converting or scaling each of the raw data inputs to create a processed data input value for each raw data input.
13. The method of claim 11 , wherein, for one or more raw data inputs, the processing of the raw data inputs comprises scaling a processed data input value for each of the raw data inputs in accordance with one or more statistical observations of the plurality of raw data inputs.
14. The method of claim 11 , wherein, for one or more sets of raw data inputs, the processing of the raw data inputs comprises deriving a processed data input value for each plurality of raw data inputs in accordance with one or more statistical observations of the plurality of raw data inputs.
15. The method of claim 11 , wherein the labeled data instances are generated with data cleaning based on one or more cleaning rules indicated by the configuration.
16. The method of claim 15 , wherein the data cleaning comprises one or more of:
identifying duplicate labeled data instances in the labeled data set;
removing the identified duplicate labeled data instances from the labeled data set;
verifying data is valid from the labeled data set;
monitoring for mandatory data in the labeled data set;
detecting conflicts with data instances in the labeled data set; and
informing the repository of the identified duplicate labeled data instances.
17. The method of claim 11 , wherein:
the configuration comprises an output time requirement parameter; and
the operations further comprise acquiring an expected data output in accordance with the output time requirement parameter.
18. The method of claim 11 , wherein the pre-processing comprises one or more of: measurement unit conversion, data type conversion, or data aggregation.
19. The method of claim 18 , wherein the data aggregation comprises one or more of: a sum, an average, a minimum, a maximum, or a count.
20. The method of claim 11 , wherein the data transformation comprises one or more of: normalization, standardization, or binning.