Git Product home page Git Product logo

Comments (3)

leewyang avatar leewyang commented on May 24, 2024

Unfortunately, Java API support in TF has been spotty with deprecation warnings and no API stability guarantees. We initially tried to support Java when the API was updated regularly with each TF release, but even then, it was mostly geared towards inference and not training. That said, contributions are always welcome!

from tensorflowonspark.

mullerhai avatar mullerhai commented on May 24, 2024

Unfortunately, Java API support in TF has been spotty with deprecation warnings and no API stability guarantees. We initially tried to support Java when the API was updated regularly with each TF release, but even then, it was mostly geared towards inference and not training. That said, contributions are always welcome!

for me when I read our source code ,I was inspired by these scala code in TFModel, we do the model implement spark Model Interface api, tensor convert df and df convert to tensor ,and invoke tensorflow session, and get the distribute partition block ,mapPartition do model train,collect all partitions result for one model

  override def transform(dataset: Dataset[_]): DataFrame = {
    val spark = dataset.sparkSession

    val inputColumns = this.getInputMapping.keys.toSeq
    val inputTensorNames = this.getInputMapping.values
    val outputTensorNames = this.getOutputMapping.keys.toSeq

    val inputDF = dataset.select(inputColumns.head, inputColumns.tail: _*)
    val inputSchema = inputDF.schema
    val outputSchema = transformSchema(inputSchema)

    val outputRDD = inputDF.rdd.mapPartitions { iter: Iterator[Row] =>
      if (TFModel.model == null || TFModel.modelDir != this.getModel) {
        // load model into a per-executor singleton reference, if needed.
        TFModel.modelDir = this.getModel
        TFModel.model = SavedModelBundle.load(this.getModel, this.getTag)
        TFModel.graph = TFModel.model.graph
        TFModel.sess = TFModel.model.session
      }

      iter.grouped(this.getBatchSize).flatMap { batch =>
        // get input batch of Rows and convert to list of input Tensors
        val inputTensors = batch2tensors(batch, inputSchema)

        var runner = TFModel.sess.runner()

        // feed input tensors
        for ((name, tensor) <- inputTensors) {
          runner = runner.feed(this.getInputMapping(name), tensor)
        }
        // fetch output tensors
        for (name <- outputTensorNames) {
          runner = runner.fetch(name)
        }

        // run the graph
        val outputTensors = runner.run()

        assert(outputTensors.map(_.shape).map(s => if (s.isEmpty) 0L else s.apply(0)).distinct.size == 1,
          "Cardinality of output tensors must match")

        // convert the list of output Tensors to a batch of output Rows
        tensors2batch(outputTensors)
      }
    }

    spark.createDataFrame(outputRDD, outputSchema)
  }

from tensorflowonspark.

mokundong avatar mokundong commented on May 24, 2024

got an error "Could not find SavedModel .pb" when submit on yarn cluster at code "TFModel.model = SavedModelBundle.load(this.getModel, this.getTag)"

from tensorflowonspark.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.