Add keras model for image classification example and api change (#4098)

scala/dllib/src/main/scala/com/intel/analytics/bigdl/dllib/example/keras/ImageClassification.scala

@@ -0,0 +1,84 @@
+/*
+ * Copyright 2016 The BigDL Authors.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.intel.analytics.bigdl.dllib.example.keras
+
+import com.intel.analytics.bigdl.dllib.NNContext
+import com.intel.analytics.bigdl.dllib.feature.image.ImageChannelNormalize
+import com.intel.analytics.bigdl.dllib.nnframes.NNImageReader
+import com.intel.analytics.bigdl.dllib.keras.layers._
+import com.intel.analytics.bigdl.dllib.utils.Shape
+import com.intel.analytics.bigdl.dllib.keras.Sequential
+import com.intel.analytics.bigdl.dllib.keras.objectives.BinaryCrossEntropy
+import com.intel.analytics.bigdl.dllib.optim._
+import com.intel.analytics.bigdl.dllib.models.lenet.Utils._
+import com.intel.analytics.bigdl.numeric.NumericFloat
+import org.apache.hadoop.fs.Path
+import org.apache.spark.sql.Row
+import org.apache.spark.sql.functions._
+
+object ImageClassification {
+  def buildMode(inputShape: Shape): Sequential[Float] = {
+    import com.intel.analytics.bigdl.numeric.NumericFloat
+    val model = Sequential()
+    model.add(Conv2D(32, 3, 3, inputShape = inputShape))
+    model.add(Activation("relu"))
+    model.add(MaxPooling2D(poolSize = (2, 2)))
+
+    model.add(Conv2D(32, 3, 3))
+    model.add(Activation("relu"))
+    model.add(MaxPooling2D(poolSize = (2, 2)))
+
+    model.add(Conv2D(64, 3, 3))
+    model.add(Activation("relu"))
+    model.add(MaxPooling2D(poolSize = (2, 2)))
+
+    model.add(Flatten())
+    model.add(Dense(64))
+    model.add(Activation("relu"))
+    model.add(Dropout(0.5))
+    model.add(Dense(1))
+    model.add(Activation("sigmoid"))
+
+    return model
+  }
+
+  def main(args: Array[String]): Unit = {
+    trainParser.parse(args, new TrainParams()).map(param => {
+      val sc = NNContext.initNNContext()
+
+      val createLabel = udf { row: Row =>
+        if (new Path(row.getString(0)).getName.contains("cat")) 1 else 2
+      }
+      val imgDF = NNImageReader.readImages(param.folder, sc, resizeH = 150, resizeW = 150)
+        .withColumn("label", createLabel(col("image")))
+      val Array(validationDF, trainingDF) = imgDF.randomSplit(Array(0.1, 0.9), seed = 42L)
+
+      val transformers = ImageChannelNormalize(0, 0, 0, 255, 255, 255)
+      val model = buildMode(Shape(3, 150, 150))
+
+      val optimMethod = new RMSprop[Float]()
+
+      model.compile(optimizer = optimMethod,
+        loss = BinaryCrossEntropy[Float](),
+        metrics = List(new Top1Accuracy[Float]()))
+      model.fit(trainingDF, batchSize = param.batchSize, nbEpoch = param.maxEpoch,
+        labelCols = Array("label"), transform = transformers, valX = validationDF)
+
+      sc.stop()
+    })
+  }
+}

scala/dllib/src/main/scala/com/intel/analytics/bigdl/dllib/keras/models/Topology.scala

@@ -577,15 +577,16 @@ abstract class KerasNet[T](implicit val tag: ClassTag[T], implicit val ev: Tenso
    x: DataFrame,
    batchSize: Int,
    nbEpoch: Int,
-   labelCol: String,
+   labelCols: Array[String],
    transform: ImageProcessing,
    valX: DataFrame)(implicit ev: TensorNumeric[T]): Unit = {
-    val trainData = df2ImageSet(x, labelCol, transform)
+    require(labelCols.length == 1, "current only support one label for dataframe of image")
+    val trainData = df2ImageSet(x, labelCols.head, transform)
     val transformer2 = ImageMatToTensor[Float]() -> ImageSetToSample[Float]()
     trainData.transform(transformer2)
 
     val valData = if (valX != null) {
-      val valSet = df2ImageSet(valX, labelCol, transform)
+      val valSet = df2ImageSet(valX, labelCols.head, transform)
       valSet.transform(transformer2)
       valSet
     } else null
@@ -597,9 +598,9 @@ abstract class KerasNet[T](implicit val tag: ClassTag[T], implicit val ev: Tenso
      x: DataFrame,
      batchSize: Int,
      nbEpoch: Int,
-     labelCol: String,
+     labelCols: Array[String],
      transform: ImageProcessing)(implicit ev: TensorNumeric[T]): Unit = {
-    this.fit(x, batchSize, nbEpoch, labelCol, transform, null)
+    this.fit(x, batchSize, nbEpoch, labelCols, transform, null)
   }
 
   /**

scala/dllib/src/test/scala/com/intel/analytics/bigdl/dllib/keras/models/TrainingSpec.scala

@@ -17,7 +17,7 @@
 package com.intel.analytics.bigdl.dllib.keras.models
 
 import com.intel.analytics.bigdl.dllib.feature.dataset.{LocalDataSet, MiniBatch, Sample}
-import com.intel.analytics.bigdl.dllib.optim.{Loss, SGD, Top1Accuracy, Top5Accuracy}
+import com.intel.analytics.bigdl.dllib.optim._
 import com.intel.analytics.bigdl.dllib.utils.python.api.PythonBigDL
 import com.intel.analytics.bigdl.dllib.tensor.TensorNumericMath.TensorNumeric
 import com.intel.analytics.bigdl.dllib.tensor.{Storage, Tensor}
@@ -32,7 +32,7 @@ import com.intel.analytics.bigdl.dllib.keras.{Sequential, ZooSpecHelper}
 import com.intel.analytics.bigdl.dllib.keras.layers._
 import com.intel.analytics.bigdl.dllib.keras.models.Sequential
 import com.intel.analytics.bigdl.dllib.keras.models.Model
-import com.intel.analytics.bigdl.dllib.keras.objectives.ZooClassNLLCriterion
+import com.intel.analytics.bigdl.dllib.keras.objectives.{BinaryCrossEntropy, ZooClassNLLCriterion}
 import com.intel.analytics.bigdl.dllib.keras.python.PythonZooKeras
 import com.intel.analytics.bigdl.dllib.nn.{CosineEmbeddingCriterion, MSECriterion, MarginRankingCriterion, ParallelCriterion}
 import com.intel.analytics.bigdl.dllib.nnframes.{NNEstimatorSpec, NNImageReader}
@@ -449,7 +449,8 @@ class TrainingSpec extends ZooSpecHelper {
     model.add(Reshape[Float](Array(169)))
     model.add(Dense[Float](2, activation = "log_softmax"))
     model.compile(optimizer = new SGD[Float](), loss = ZooClassNLLCriterion[Float]())
-    model.fit(imgDF, batchSize = 1, nbEpoch = 1, labelCol = "label", transform = transformers)
+    model.fit(imgDF, batchSize = 1, nbEpoch = 1, labelCols = Array("label"),
+      transform = transformers)
     val predDf = model.predict(imgDF, predictionCol = "predict", transform = transformers)
     predDf.show()
     model.evaluate(imgDF, batchSize = 1, labelCol = "label", transform = transformers)