Hi Jim, It is possible to use raw artificial neural networks by means of FeedForwardTrainer. It is [ml] package private, so your code should be in that package too. Basically, you need to do the same as it is done in MultilayerPerceptronClassifier but without encoding the output as one-hot: https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala#L168 Here is the sketch of the code if your outputs are continuous and are within [0, 1] interval: val topology = FeedForwardTopology.multiLayerPerceptron(layers, useSoftMax = false) val FeedForwardTrainer = new FeedForwardTrainer(topology, layers(0), layers.last) val mlpModel = FeedForwardTrainer.train(data)
For the outputs in that are in the range of (-infinity, +infinity), the new loss function is needed. It is implemented but not merged in Spark yet, unfortunately: https://github.com/avulanov/spark/tree/autoencoder-mlp Best regards, Alexander -----Original Message----- From: Jim Carroll [mailto:jimfcarr...@gmail.com] Sent: Sunday, March 27, 2016 6:25 AM To: user@spark.apache.org Subject: Non-classification neural networks Hello all, We were using the old "Artificial Neural Network" : https://github.com/apache/spark/pull/1290 This code appears to have been incorporated in 1.5.2 but it's only exposed publicly via the MultilayerPerceptronClassifier. Is there a way to use the old feedforward/backprop non-classification functionality? It appears to be buried in private classes and it's not obvious to me if the MultilayerPerceptronClassifier can be used without the classification. The doc says "Number of outputs has to be equal to the total number of labels." What if the output is continuous and you want to simply do prediction? Thanks Jim -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/Non-classification-neural-networks-tp26604.html Sent from the Apache Spark User List mailing list archive at Nabble.com. --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
