From c1cf09c8f38adcc12d31fdc18070ee07564be860 Mon Sep 17 00:00:00 2001
From: Song Jiaming <litchilitchy@qq.com>
Date: Mon, 15 Jun 2020 16:25:34 +0800
Subject: [PATCH] Infernce Model string support of TFNet (#2452)

Noted InferenceModel predict batchSize is removed
---
 .../bigdl/dllib/inference/InferenceModel.scala  | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/spark/dl/src/main/scala/com/intel/analytics/bigdl/dllib/inference/InferenceModel.scala b/spark/dl/src/main/scala/com/intel/analytics/bigdl/dllib/inference/InferenceModel.scala
index d29dc7515e1..b3b4eb273c6 100644
--- a/spark/dl/src/main/scala/com/intel/analytics/bigdl/dllib/inference/InferenceModel.scala
+++ b/spark/dl/src/main/scala/com/intel/analytics/bigdl/dllib/inference/InferenceModel.scala
@@ -24,9 +24,11 @@ import java.util.{List => JList}
 
 import com.intel.analytics.bigdl.nn.abstractnn.Activity
 import com.intel.analytics.bigdl.tensor.Tensor
+import com.intel.analytics.bigdl.tensor.TensorNumericMath.TensorNumeric
 import com.sun.xml.internal.bind.v2.TODO
 
 import scala.collection.JavaConverters._
+import scala.reflect.ClassTag
 
 class InferenceModel(private var autoScalingEnabled: Boolean = true,
                      private var concurrentNum: Int = 20,
@@ -37,7 +39,6 @@ class InferenceModel(private var autoScalingEnabled: Boolean = true,
 
   require(concurrentNum > 0, "concurrentNum should > 0")
 
-  private var batchCnt: Int = 0
   @transient var inferenceSummary: InferenceSummary = null
   /**
    * default constructor, will create a InferenceModel with auto-scaling enabled.
@@ -743,17 +744,17 @@ class InferenceModel(private var autoScalingEnabled: Boolean = true,
     val model: AbstractModel = retrieveModel()
     try {
       val begin = System.nanoTime()
-      val batchSize = if (inputActivity.isTensor) {
-        inputActivity.toTensor[Float].size(1)
-      } else {
-        val sampleKey = inputActivity.toTable.keySet.head
-        inputActivity.toTable(sampleKey).asInstanceOf[Tensor[Float]].size(1)
-      }
+//      val batchSize = if (inputActivity.isTensor) {
+//        inputActivity.toTensor[T].size(1)
+//      } else {
+//        val sampleKey = inputActivity.toTable.keySet.head
+//        inputActivity.toTable(sampleKey).asInstanceOf[Tensor[T]].size(1)
+//      }
       val result = model.predict(inputActivity)
       val end = System.nanoTime()
 
       val latency = end - begin
-      val name = s"model predict for batch ${batchSize}"
+      val name = s"model predict for batch"
       InferenceSupportive.logger.info(s"$name time elapsed [${latency/1e9} s, ${latency/1e6} ms].")
 
       result