The Machine Learning Extractor is a data extraction tool using machine learning models in order to identify and report on data targeted for data extraction.
This activity is the companion of UiPath Document Understanding Models, as the means to consume such models within your workflows.
The ML approach is strongly recommended for structured or semi-structured documents in which layouts of different document providers vary greatly. Given its machine learning approach, the extractor uses a trained machine learning model, that learns and can then infer values for the targeted fields, even from documents and layouts it has never seem before. In other words, if documents do not follow a text or layout pattern, the Machine Learning Extractor may be a good option for your use case.
The Machine Learning Model can be used in multiple ways:
- with one of UiPath's public Document Understanding endpoints, if you wish to use generic models targeting certain document types; or
- with custom trained machine learning models starting from the UiPath Document Understanding available models.
This extractor can be trained / re-trained. See the How to Train section for details.
Images with a resolution lower than 50 x 50 pixels cannot be processed, generating an error.
You need to use
- one of UiPath's public Document Understanding endpoints for data extraction, or
- machine learning models hosted in AI Center in Automation Cloud, or
- machine learning models hosted in AI Center on-prem, but licensed through Automation Cloud, you need to use your Automation Cloud Document Understanding API Key.
To use the Machine Learning Extractor with on-prem licensing, you need to host your Document Understanding models in your AI Center on-prem (air-gapped install) instance.
If the endpoint you are using is licensed through Automation Cloud, you need to provide your Cloud Document Understanding API Key.
If you are using the Machine Learning Extractor with either a UiPath Document Understanding public endpoint, or with a public ML Skill in AI Center, then you need to configure the Endpoint argument of the activity with the corresponding URL.
If you are using the Machine Learning Extractor with a deployed ML Skill, then you need to configure the ML Skill argument of the activity with the correct selection from your AI Center hosted ML skills list.
If you try to set both options, an error will be displayed - either in the Configuration Wizard, or in the workflow directly:
When first dropped in a Data Extraction Scope, the Machine Learning Extractor will open a configuration wizard. The same wizard is available if you open the Configure Extractors wizard of the Data Extraction Scope and click on the configuration icon under the extractor's name.
The wizard allows you to enter either an Endpoint or an ML Skill, as well as provide an ApiKey (if necessary). If you enter an Endpoint and an ApiKey, you need to enter them without quotes - and the values cannot be variables.
If you choose, you can use the "Update Activity Arguments" option to pre-populate the activity arguments with the values added in the wizard.
When clicking on the "Get Capabilities" option, the Machine Learning Extractor will "read and report" on its internal capabilities (what document types and what fields it knows how to process), with the purpose of helping you configure data extraction correctly.
It is recommended you use the ML Extractor Capabilities wizard every time that you change the ML Skill or Endpoint used in your workflow, to thus ensure the configuration and taxonomy mapping in the Data Extraction Scope remains valid.
Once the ML Extractor Capabilities wizard is run, you will notice that the Configure Extractors wizard does not present text boxes for taxonomy mapping anymore, but drop-down lists.
Expand the document type that you want to extract data for, and start selecting the fields you are targeting, by checking the checkboxes next to the appropriate fields and by selecting, from the available drop-down list, the correct field from the ML model you wish to map to each particular field. The drop-down list contains all fields that the Machine Learning Extractor, using the endpoint entered in the Machine Learning Extractor wizard, declares as extraction capability.
To check if you are using the latest capabilities of the extractor, you can click the Get or refresh extractor capabilities which opens the Machine Learning Extractor wizard.
You cannot choose the same option for two distinct fields.
If you wish to use the Extractor's training capabilities as well, it is strongly recommended that you enter a unique string, alphanumeric value, in the Framework Alias configuration field, and then use the exact same string value in the corresponding Framework Alias field of Train Extractors Scope configuration, for the trainers that need to receive the complete training data.
Select the Save button once all data is configured properly.
Use the Machine Learning Extractor Trainer activity within a Train Extractors Scope, in order to collect training data for your Machine Learning Extractor model instance. The thus collected data can be used for curation and then importing with the purpose of training, in your instance of AI Center (Cloud or on-premises).
Updated about a month ago