Add QA sweepable estimator in AutoML (#6781)

* Add QA sweepable

* clean

Author: zewditu

src/Microsoft.ML.AutoML/CodeGen/estimator-schema.json

@@ -73,7 +73,8 @@
               "ForecastBySsa",
               "TextClassifcation",
               "SentenceSimilarity",
-              "ObjectDetection"
+              "ObjectDetection",
+              "QuestionAnswering"
             ]
           },
           "nugetDependencies": {

src/Microsoft.ML.AutoML/CodeGen/question_answering_search_space.json

@@ -0,0 +1,56 @@
+{
+  "$schema": "./search-space-schema.json#",
+  "name": "question_answering_option",
+  "search_space": [
+    {
+      "name": "ContextColumnName",
+      "type": "string",
+      "default": "Context"
+    },
+    {
+      "name": "QuestionColumnName",
+      "type": "string",
+      "default": "Question"
+    },
+    {
+      "name": "TrainingAnswerColumnName",
+      "type": "string",
+      "default": "TrainingAnswer"
+    },
+    {
+      "name": "AnswerIndexStartColumnName",
+      "type": "string",
+      "default": "AnswerStart"
+    },
+    {
+      "name": "ScoreColumnName",
+      "type": "string",
+      "default": "Score"
+    },
+    {
+      "name": "predictedAnswerColumnName",
+      "type": "string",
+      "default": "Answer"
+    },
+    {
+      "name": "BatchSize",
+      "type": "integer",
+      "default": 4
+    },
+    {
+      "name": "MaxEpochs",
+      "type": "integer",
+      "default": 10
+    },
+    {
+      "name": "TopKAnswers",
+      "type": "integer",
+      "default": 3
+    },
+    {
+      "name": "Architecture",
+      "type": "bertArchitecture",
+      "default": "BertArchitecture.Roberta"
+    }
+  ]
+}

src/Microsoft.ML.AutoML/CodeGen/search-space-schema.json

@@ -146,7 +146,8 @@
         "dnn_featurizer_image_option",
         "text_classification_option",
         "sentence_similarity_option",
-        "object_detection_option"
+        "object_detection_option",
+        "question_answering_option"
       ]
     },
     "option_name": {
@@ -210,7 +211,13 @@
         "Steps",
         "MaxEpoch",
         "InitLearningRate",
-        "WeightDecay"
+        "WeightDecay",
+        "ContextColumnName",
+        "QuestionColumnName",
+        "TrainingAnswerColumnName",
+        "AnswerIndexStartColumnName",
+        "predictedAnswerColumnName",
+        "TopKAnswers"
       ]
     },
     "option_type": {

src/Microsoft.ML.AutoML/CodeGen/trainer-estimators.json

@@ -532,6 +532,13 @@
       "usingStatements": [ "Microsoft.ML", "Microsoft.ML.Trainers", "Microsoft.ML.TorchSharp" ],
       "searchOption": "object_detection_option"
     },
+    {
+      "functionName": "QuestionAnswering",
+      "estimatorTypes": [ "MultiClassification" ],
+      "nugetDependencies": [ "Microsoft.ML", "Microsoft.ML.TorchSharp" ],
+      "usingStatements": [ "Microsoft.ML", "Microsoft.ML.Trainers", "Microsoft.ML.TorchSharp" ],
+      "searchOption": "question_answering_option"
+    },
     {
       "functionName": "ForecastBySsa",
       "estimatorTypes": [ "Forecasting" ],

src/Microsoft.ML.AutoML/Microsoft.ML.AutoML.csproj

@@ -69,7 +69,7 @@
     <AdditionalFiles Include="CodeGen\code_gen_flag.json" />
     <AdditionalFiles Include="CodeGen\*-estimators.json" />
   </ItemGroup>
-
+  
   <ItemGroup>
     <EmbeddedResource Include="Tuner\Portfolios.json">
       <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>

src/Microsoft.ML.AutoML/SweepableEstimator/Estimators/QuestionAnswering.cs

@@ -0,0 +1,28 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+using Microsoft.ML.TorchSharp;
+using Microsoft.ML.TorchSharp.NasBert;
+using Microsoft.ML.TorchSharp.Roberta;
+
+namespace Microsoft.ML.AutoML.CodeGen
+{
+    internal partial class QuestionAnsweringMulti
+    {
+        public override IEstimator<ITransformer> BuildFromOption(MLContext context, QuestionAnsweringOption param)
+        {
+            return context.MulticlassClassification.Trainers.QuestionAnswer(
+                   contextColumnName: param.ContextColumnName,
+                   questionColumnName: param.QuestionColumnName,
+                   trainingAnswerColumnName: param.TrainingAnswerColumnName,
+                   answerIndexColumnName: param.AnswerIndexStartColumnName,
+                   predictedAnswerColumnName: param.PredictedAnswerColumnName,
+                   scoreColumnName: param.ScoreColumnName,
+                   batchSize: param.BatchSize,
+                   maxEpochs: param.MaxEpochs,
+                   topK: param.TopKAnswers,
+                   architecture: BertArchitecture.Roberta);
+        }
+
+    }
+}