fix: fix issues with form recognizer parsing and form ontology learner

cognitive/src/main/scala/com/microsoft/azure/synapse/ml/cognitive/FormOntologyLearner.scala

@@ -46,7 +46,10 @@ class FormOntologyLearner(override val uid: String) extends Estimator[FormOntolo
   def this() = this(Identifiable.randomUID("FormOntologyLearner"))
 
   private[ml] def extractOntology(fromRow: Row => AnalyzeResponse)(r: Row): StructType = {
-    val fieldResults = fromRow(r.getStruct(0)).analyzeResult.documentResults.get.head.fields
+    val fieldResults = fromRow(r.getStruct(0)).analyzeResult.documentResults
+      .getOrElse(throw new IllegalArgumentException("A row does not have a `analyzeResult.documentResults` field," +
+        " please filter these out before using the FormOntologyLearner"))
+      .head.fields
     new StructType(fieldResults
       .mapValues(_.toFieldResultRecursive.toSimplifiedDataType)
       .map({ case (name, dt) => StructField(name, dt) }).toArray)

cognitive/src/main/scala/com/microsoft/azure/synapse/ml/cognitive/FormRecognizer.scala

@@ -18,7 +18,7 @@ import spray.json._
 
 abstract class FormRecognizerBase(override val uid: String) extends CognitiveServicesBaseNoHandler(uid)
   with HasCognitiveServiceInput with HasInternalJsonOutputParser with BasicAsyncReply
-  with HasImageInput with HasSetLocation with HasSetLinkedService {
+  with HasImageInput with HasSetLocation with HasSetLinkedService with HasModelVersion {
 
   override protected def prepareEntity: Row => Option[AbstractHttpEntity] = {
     r =>
@@ -306,21 +306,8 @@ class GetCustomModel(override val uid: String) extends CognitiveServicesBase(uid
 
   def urlPath: String = "formrecognizer/v2.1/custom/models"
 
-  override protected def prepareUrl: Row => String = {
-    val urlParams: Array[ServiceParam[Any]] =
-      getUrlParams.asInstanceOf[Array[ServiceParam[Any]]];
-    // This semicolon is needed to avoid argument confusion
-    { row: Row =>
-      val base = getUrl + s"/${getValue(row, modelId)}"
-      val appended = if (!urlParams.isEmpty) {
-        "?" + URLEncodingUtils.format(urlParams.flatMap(p =>
-          getValueOpt(row, p).map(v => p.name -> p.toValueString(v))
-        ).toMap)
-      } else {
-        ""
-      }
-      base + appended
-    }
+  override protected def prepareUrlRoot: Row => String = { row =>
+    getUrl + s"/${getValue(row, modelId)}"
   }
 
   override protected def prepareMethod(): HttpRequestBase = new HttpGet()
@@ -347,21 +334,8 @@ class AnalyzeCustomModel(override val uid: String) extends FormRecognizerBase(ui
 
   def urlPath: String = "formrecognizer/v2.1/custom/models"
 
-  override protected def prepareUrl: Row => String = {
-    val urlParams: Array[ServiceParam[Any]] =
-      getUrlParams.asInstanceOf[Array[ServiceParam[Any]]];
-    // This semicolon is needed to avoid argument confusion
-    { row: Row =>
-      val base = getUrl + s"/${getValue(row, modelId)}/analyze"
-      val appended = if (!urlParams.isEmpty) {
-        "?" + URLEncodingUtils.format(urlParams.flatMap(p =>
-          getValueOpt(row, p).map(v => p.name -> p.toValueString(v))
-        ).toMap)
-      } else {
-        ""
-      }
-      base + appended
-    }
+  override protected def prepareUrlRoot: Row => String = {row =>
+    getUrl + s"/${getValue(row, modelId)}/analyze"
   }
 
   override protected def responseDataType: DataType = AnalyzeResponse.schema

cognitive/src/main/scala/com/microsoft/azure/synapse/ml/cognitive/FormRecognizerV3.scala

@@ -34,20 +34,19 @@ trait HasPrebuiltModelID extends HasServiceParams {
   def getPrebuiltModelIdCol: String = getVectorParam(prebuiltModelId)
 }
 
-object AnalyzeDocument extends ComplexParamsReadable[AnalyzeDocument] {
-  // Different versions might have different results so make sure tests pass before updating
-  val DefaultAPIVersion = "2022-01-30-preview"
-}
+object AnalyzeDocument extends ComplexParamsReadable[AnalyzeDocument]
 
 class AnalyzeDocument(override val uid: String) extends CognitiveServicesBaseNoHandler(uid)
   with HasCognitiveServiceInput with HasInternalJsonOutputParser with BasicAsyncReply
-  with HasPrebuiltModelID with HasPages with HasLocale
+  with HasPrebuiltModelID with HasPages with HasLocale with HasAPIVersion
   with HasImageInput with HasSetLocation with BasicLogging {
   logClass()
 
+  setDefault(apiVersion -> Left("2022-01-30-preview"))
+
   def this() = this(Identifiable.randomUID("AnalyzeDocument"))
 
-  def urlPath: String = "formrecognizer/documentModels/"
+  def urlPath: String = "formrecognizer/documentModels"
 
   val stringIndexType = new ServiceParam[String](this, "stringIndexType", "Method used to " +
     "compute string offset and length.", {
@@ -75,21 +74,8 @@ class AnalyzeDocument(override val uid: String) extends CognitiveServicesBaseNoH
         "Payload needs to contain image bytes or url. This code should not run"))
   }
 
-  override protected def prepareUrl: Row => String = {
-    val urlParams: Array[ServiceParam[Any]] =
-      getUrlParams.asInstanceOf[Array[ServiceParam[Any]]];
-    // This semicolon is needed to avoid argument confusion
-    { row: Row =>
-      val base = getUrl + s"${getValue(row, prebuiltModelId)}:analyze?api-version=${AnalyzeDocument.DefaultAPIVersion}"
-      val appended = if (!urlParams.isEmpty) {
-        "&" + URLEncodingUtils.format(urlParams.flatMap(p =>
-          getValueOpt(row, p).map(v => p.name -> p.toValueString(v))
-        ).toMap)
-      } else {
-        ""
-      }
-      base + appended
-    }
+  override protected def prepareUrlRoot: Row => String = { row =>
+    getUrl + s"/${getValue(row, prebuiltModelId)}:analyze"
   }
 
 }

cognitive/src/main/scala/com/microsoft/azure/synapse/ml/cognitive/FormRecognizerV3Schemas.scala

@@ -27,8 +27,8 @@ case class AnalyzeResultV3(apiVersion: String,
 
 case class PageResultV3(pageNumber: Int,
                         angle: Double,
-                        width: Int,
-                        height: Int,
+                        width: Double,
+                        height: Double,
                         unit: String,
                         spans: Option[Seq[FormSpan]],
                         words: Option[Seq[FormWord]],
@@ -37,19 +37,19 @@ case class PageResultV3(pageNumber: Int,
 
 case class FormSpan(offset: Int, length: Int)
 
-case class FormWord(content: String, boundingBox: Option[Seq[Int]], confidence: Double, span: FormSpan)
+case class FormWord(content: String, boundingBox: Option[Seq[Double]], confidence: Double, span: FormSpan)
 
-case class FormSelectionMark(state: String, boundingBox: Option[Seq[Int]], confidence: Double, span: FormSpan)
+case class FormSelectionMark(state: String, boundingBox: Option[Seq[Double]], confidence: Double, span: FormSpan)
 
-case class FormLine(content: String, boundingBox: Option[Seq[Int]], spans: Option[Seq[FormSpan]])
+case class FormLine(content: String, boundingBox: Option[Seq[Double]], spans: Option[Seq[FormSpan]])
 
 case class TableResultV3(rowCount: Int,
                          columnCount: Int,
                          boundingRegions: Option[Seq[BoundingRegion]],
                          spans: Option[Seq[FormSpan]],
                          cells: Option[Seq[FormCell]])
 
-case class BoundingRegion(pageNumber: Int, boundingBox: Option[Seq[Int]])
+case class BoundingRegion(pageNumber: Int, boundingBox: Option[Seq[Double]])
 
 case class FormCell(kind: String,
                     rowIndex: Int,

cognitive/src/test/scala/com/microsoft/azure/synapse/ml/cognitive/split1/FormOntologyLearnerSuite.scala

@@ -5,6 +5,7 @@ package com.microsoft.azure.synapse.ml.cognitive.split1
 
 import com.microsoft.azure.synapse.ml.cognitive._
 import com.microsoft.azure.synapse.ml.core.test.fuzzing.{EstimatorFuzzing, TestObject}
+import org.apache.spark.SparkException
 import org.apache.spark.ml.util.MLReadable
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.{ArrayType, DoubleType, StringType, StructType}
@@ -31,8 +32,30 @@ class FormOntologyLearnerSuite extends EstimatorFuzzing[FormOntologyLearner] wit
     "https://mmlsparkdemo.blob.core.windows.net/ignite2021/forms/2009/Invoice12241.pdf"
   ).toDF("url")
 
+  lazy val tableUrlDF: DataFrame = Seq(
+    "https://mmlspark.blob.core.windows.net/datasets/FormRecognizer/tables1.pdf"
+  ).toDF("url")
+
   lazy val df: DataFrame = analyzeInvoices.transform(urlDF).cache()
 
+  test("Yields a reasonable error message when input rows dont contain documentResults") {
+    val analyzedDf = new AnalyzeLayout()
+      .setSubscriptionKey(cognitiveKey)
+      .setLocation("eastus")
+      .setImageUrlCol("url")
+      .setOutputCol("layout")
+      .setConcurrency(5)
+      .transform(tableUrlDF)
+
+    assertThrows[SparkException] {
+      new FormOntologyLearner()
+        .setInputCol("layout")
+        .setOutputCol("unified_ontology")
+        .fit(analyzedDf)
+        .transform(analyzedDf)
+    }
+  }
+
   test("Basic Usage") {
 
     val targetSchema = new StructType()

cognitive/src/test/scala/com/microsoft/azure/synapse/ml/cognitive/split1/FormRecognizerSuite.scala

@@ -130,6 +130,8 @@ trait FormRecognizerUtils extends TestBase with CognitiveKey with Flaky {
 
   lazy val bytesDF5: DataFrame = createTestDataframe(baseUrl, Seq("id1.jpg"), returnBytes = true)
 
+  lazy val imageDf6: DataFrame = createTestDataframe(baseUrl, Seq("tables1.pdf"), returnBytes = false)
+
   lazy val pdfDf1: DataFrame = createTestDataframe(baseUrl, Seq("layout2.pdf"), returnBytes = false)
 
   lazy val pdfDf2: DataFrame = createTestDataframe(baseUrl, Seq("invoice1.pdf", "invoice3.pdf"), returnBytes = false)

cognitive/src/test/scala/com/microsoft/azure/synapse/ml/cognitive/split1/FormRecognizerV3Suite.scala

@@ -60,6 +60,17 @@ class AnalyzeDocumentSuite extends TransformerFuzzing[AnalyzeDocument] with Form
     super.assertDFEq(prep(df1), prep(df2))(eq)
   }
 
+  test("basic usage with tables") {
+    val fromRow = AnalyzeDocumentResponse.makeFromRowConverter
+    analyzeDocument
+      .setPrebuiltModelId("prebuilt-layout")
+      .setImageUrlCol("source")
+      .transform(imageDf6)
+      .collect()
+      .map(r => fromRow(r.getAs[Row]("result")))
+      .foreach(r => assert(r.analyzeResult.pages.get.head.pageNumber >= 0))
+  }
+
   def analyzeDocument: AnalyzeDocument = new AnalyzeDocument()
     .setSubscriptionKey(cognitiveKey)
     .setLocation("eastus")
@@ -128,7 +139,7 @@ class AnalyzeDocumentSuite extends TransformerFuzzing[AnalyzeDocument] with Form
       resultAssert(result, "WA WASHINGTON\n20 1234567XX1101\nDRIVER LICENSE\nFEDERAL LIMITS APPLY\n" +
         "4d LIC#WDLABCD456DG 9CLASS\nDONORS\n1 TALBOT\n2 LIAM R.\n3 DOB 01/06/1958\n",
         "Address,CountryRegion,DateOfBirth,DateOfExpiration,DocumentNumber," +
-        "Endorsements,FirstName,LastName,Region,Restrictions,Sex")
+          "Endorsements,FirstName,LastName,Region,Restrictions,Sex")
     }
   }