Form Recognizer Javascript SDK demo doesn't match the samples in the getting started guide.

Question

Form Recognizer Javascript SDK demo doesn't match the samples in the getting started guide.

counterpoint2022 36

I followed the getting started guide.

And the General document model output doesn't match what is described in the guide. It seems the table extraction portion of the code is missing.
Here is my output:

Key-Value Pairs:

Key : "QUARTERLY REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934"
Value: ":selected:" (0.76)
Key : "For the Quarterly Period Ended"
Value: "2020" (0.155)
Key : "March 31,"
Value: "<undefined>" (0.118)
Key : "TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d) OF THE SECURITIES EXCHANGE ACT OF 1934"
Value: ":unselected:" (0.761)
Key : "For the Transition Period From"
Value: "<undefined>" (0.227)
Key : "to"
Value: "<undefined>" (0.267)
Key : "Commission File Number"
Value: "001-37845" (0.863)
Key : "(STATE OF INCORPORATION)"
Value: "WASHINGTON
(425) 882-8080" (0.38)
Key : "(I.R.S. ID)"
Value: "91-1144442" (0.888)
Key : "Securities registered pursuant to Section 12(g) of the Act:"
Value: "NONE" (0.816)
Key : "Yes"
Value: ":selected:" (0.911)
Key : "No"
Value: ":unselected:" (0.911)
Key : "Yes"
Value: ":selected:" (0.911)
Key : "No"
Value: ":unselected:" (0.911)
Key : "Large accelerated filer"
Value: ":selected:" (0.874)
Key : "Accelerated filer"
Value: ":unselected:" (0.889)
Key : "Non-accelerated filer"
Value: ":unselected:" (0.889)
Key : "Smaller reporting company"
Value: ":unselected:" (0.874)
Key : "Emerging growth company"
Value: ":unselected:" (0.874)
Key : "If an emerging growth company, indicate by check mark if the registrant has elected not to use the extended transition period for complying with any new or revised financial accounting standards provided pursuant to Section 13(a) of the Exchange Act."
Value: ":unselected:" (0.674)
Key : "Yes"
Value: ":unselected:" (0.911)
Key : "No"
Value: ":selected:" (0.911)

Did anyone run into the same problem? Thanks

Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2022-07-25T11:25:24.747+00:00

@counterpoint2022 I think the quickstart is not handling the complete response from the general document or the "prebuilt-document" parameter passed to beginAnalyzeDocument() method. It is only prininting the key value pairs from the actual response received.
For a quick lookup if you use the studio and run the same document through the general document analysis page the complete response can be seen as a JSON format and the table data is available along with the key value pairs and other fields of the analysis response.
counterpoint2022 36 Reputation points

2022-07-25T22:43:23.897+00:00

Thanks. I saw the working version in the Reco Studio before, but for my POC, I need to be able to access those information programmatically. In the getting started guide, the table extraction portion of the code seems to be missing. The expected output clearly show the "Entities" portion. Basically, the sample code the expected output do not match. Can you please ask someone provide the table extraction portion of the code?
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2022-07-26T13:17:30.267+00:00
I am a newbie with JS, but I think the sample scripts for the v4-beta version are available on the Azure JS SDK repo under the tag ai-form-recognizer_4.0.0-beta.4, there could be other versions of these tags as the preview version of SDK and API are updated.

However, to extract the complete table along with the cells there isn't a direct script available but the extractLayout.js script does provide you the sample to get the table count data to be printed.
By adding the tables section from the poller result, you can print the table metadata with this snippet. If you want to print each cell of all the tables you can further iterate and print ${table.cells[].content}

const { keyValuePairs, entities, tables } = await poller.pollUntilDone(); if (tables.length <= 0) { console.log("No tables were extracted from the document."); } else { console.log("Tables:"); for (const table of tables) { console.log( `- Extracted table: ${table.columnCount} columns, ${table.rowCount} rows (${table.cells.length} cells)` ); } } }
counterpoint2022 36 Reputation points

2022-07-26T22:54:26.573+00:00

Thank You. I am able to get to work. I have no way of knowing about the Beta 4 code without your help. It would be very helpful if someone in MS can keep the current "getting started guide" up to date.

In the examples, we are using the sample documents provided. How do I pass in my own document/image for analysis? In the example the document has URL. is it possible to load a file with an API instead of an URL?

I found the document formrecognizerclient. I will try to see if it works.

Thanks
counterpoint2022 36 Reputation points

2022-07-27T00:41:28.16+00:00

I tried to used the API described in formrecognizerclient. I am trying the load file locally instead of using a URL.

Example code:
const path = "./Invoice_7.pdf";
const readStream = fs.createReadStream(path);
const client = new FormRecognizerClient(endpoint, new AzureKeyCredential(apiKey));
const poller = await client.beginRecognizeContent(readStream, "application/pdf", {
onProgress: (state) => { console.log(status: ${state.status}); }
});
const pages = await poller.pollUntilDone();

But I think the code is also out of date. I am getting the error "An error occurred: TypeError: FormRecognizerClient is not a constructor
". Can you try to get an updated code ? Thanks
counterpoint2022 36 Reputation points

2022-07-27T06:37:22.363+00:00

I am able to get it to work but I had install both the V4Beta.4 and V3.2.0 SDKs. This was recommended by migration guide.
It seems the FormTrainingClient is only available in V3. Also, the table extraction doesn't seem to be fully working in V3. Again the sample code in the getting started guide is out of date.
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2022-07-27T12:16:05.617+00:00

The reference to the samples I mentioned is from the SDK document reference from the v4 beta page. The reference contains details about the new training client DocumentModelAdministrationClient() but it is not yet updated in the quickstart documentation. Based on my experience of previous versions of FR API's the updates in quickstart documentation is linked to github repo samples as the scripts are frequently updated with every preview SDK release. We will definitely pass this feedback to the content owners of this page to update at the earliest.

To be honest, version 4.x.x of the SDK is a full redesign of the Azure Form Recognizer client library and most of the functionality from previous 3.0.0 versions is still being ported to support the changes in the FR REST API changes for v3.0

In V3 version of SDK i.e FR REST API v2.0, the result of table extraction differed as the layout operation was part of beginRecognizeContent() method and the table data extraction feature improved in FR API v3.0 with support for multi page extraction and other improvements.