linkedin / spark-tfrecord

Read and write Tensorflow TFRecord data from Apache Spark.

Version Matrix

Spark-TFRecord

A library for reading and writing Tensorflow TFRecord data from Apache Spark. The implementation is based on Spark Tensorflow Connector, but it is rewritten in Spark FileFormat trait to provide the partitioning function.

Including the library

The artifacts are published to bintray and maven central repositories.

  • Version 0.1.x targets Spark 2.3 and Scala 2.11
  • Version 0.2.x targets Spark 2.4 and both Scala 2.11 and 2.12
  • Version 0.3.x targets Spark 3.0 and Scala 2.12

To use the package, please include the dependency as follows

<dependency>
  <groupId>com.linkedin.sparktfrecord</groupId>
  <artifactId>spark-tfrecord_2.12</artifactId>
  <version>your.version</version>
</dependency>

Building the library

The library can be built with Maven 3.3.9 or newer as shown below:

# Build Spark-TFRecord
git clone https://github.com/linkedin/spark-tfrecord.git
cd spark-tfrecord
mvn -Pscala-2.12 clean install

# One can specify the spark version and tensorflow hadoop version, for example
mvn -Pscala-2.12 clean install -Dspark.version=3.0.0 -Dtensorflow.hadoop.version=1.15.0

Using Spark Shell

Run this library in Spark using the --jars command line option in spark-shell, pyspark or spark-submit. For example:

$SPARK_HOME/bin/spark-shell --jars target/spark-tfrecord_2.12-0.3.0.jar

Features

This library allows reading TensorFlow records in local or distributed filesystem as Spark DataFrames. When reading TensorFlow records into Spark DataFrame, the API accepts several options:

  • load: input path to TensorFlow records. Similar to Spark can accept standard Hadoop globbing expressions.
  • schema: schema of TensorFlow records. Optional schema defined using Spark StructType. If not provided, the schema is inferred from TensorFlow records.
  • recordType: input format of TensorFlow records. By default it is Example. Possible values are:

When writing Spark DataFrame to TensorFlow records, the API accepts several options:

  • save: output path to TensorFlow records. Output path to TensorFlow records on local or distributed filesystem. compression. While reading compressed TensorFlow records, codec can be inferred automatically, so this option is not required for reading.
  • recordType: output format of TensorFlow records. By default it is Example. Possible values are:

The writer support partitionBy operation. So the following command will partition the output by "partitionColumn".

df.write.mode(SaveMode.Overwrite).partitionBy("partitionColumn").format("tfrecord").option("recordType", "Example").save(output_dir)

Note we use format("tfrecord") instead format("tfrecords"). So if you migrate from Spark-Tensorflow-Connector, make sure this is changed accordingly.

Schema inference

This library supports automatic schema inference when reading TensorFlow records into Spark DataFrames. Schema inference is expensive since it requires an extra pass through the data.

The schema inference rules are described in the table below:

TFRecordType Feature Type Inferred Spark Data Type
Example, SequenceExample Int64List LongType if all lists have length=1, else ArrayType(LongType)
Example, SequenceExample FloatList FloatType if all lists have length=1, else ArrayType(FloatType)
Example, SequenceExample BytesList StringType if all lists have length=1, else ArrayType(StringType)
SequenceExample FeatureList of Int64List ArrayType(ArrayType(LongType))
SequenceExample FeatureList of FloatList ArrayType(ArrayType(FloatType))
SequenceExample FeatureList of BytesList ArrayType(ArrayType(StringType))

Supported data types

The supported Spark data types are listed in the table below:

Type Spark DataTypes
Scalar IntegerType, LongType, FloatType, DoubleType, DecimalType, StringType, BinaryType
Array VectorType, ArrayType of IntegerType, LongType, FloatType, DoubleType, DecimalType, BinaryType, or StringType
Array of Arrays ArrayType of ArrayType of IntegerType, LongType, FloatType, DoubleType, DecimalType, BinaryType, or StringType

Usage Examples

Python API

TF record Import/export

Run PySpark with the spark_connector in the jars argument as shown below:

$SPARK_HOME/bin/pyspark --jars target/spark-tfrecord_2.12-0.3.0.jar

The following Python code snippet demonstrates usage on test data.

from pyspark.sql.types import *

path = "test-output.tfrecord"

fields = [StructField("id", IntegerType()), StructField("IntegerCol", IntegerType()),
          StructField("LongCol", LongType()), StructField("FloatCol", FloatType()),
          StructField("DoubleCol", DoubleType()), StructField("VectorCol", ArrayType(DoubleType(), True)),
          StructField("StringCol", StringType())]
schema = StructType(fields)
test_rows = [[11, 1, 23, 10.0, 14.0, [1.0, 2.0], "r1"], [21, 2, 24, 12.0, 15.0, [2.0, 2.0], "r2"]]
rdd = spark.sparkContext.parallelize(test_rows)
df = spark.createDataFrame(rdd, schema)
df.write.mode("overwrite").format("tfrecord").option("recordType", "Example").save(path)
df = spark.read.format("tfrecord").option("recordType", "Example").load(path)
df.show()

Scala API

Run Spark shell with the spark_connector in the jars argument as shown below:

$SPARK_HOME/bin/spark-shell --jars target/spark-tfrecord_2.12-0.3.0.jar

The following Scala code snippet demonstrates usage on test data.

import org.apache.commons.io.FileUtils
import org.apache.spark.sql.{ DataFrame, Row }
import org.apache.spark.sql.catalyst.expressions.GenericRow
import org.apache.spark.sql.types._

val path = "test-output.tfrecord"
val testRows: Array[Row] = Array(
new GenericRow(Array[Any](11, 1, 23L, 10.0F, 14.0, List(1.0, 2.0), "r1")),
new GenericRow(Array[Any](21, 2, 24L, 12.0F, 15.0, List(2.0, 2.0), "r2")))
val schema = StructType(List(StructField("id", IntegerType),
                             StructField("IntegerCol", IntegerType),
                             StructField("LongCol", LongType),
                             StructField("FloatCol", FloatType),
                             StructField("DoubleCol", DoubleType),
                             StructField("VectorCol", ArrayType(DoubleType, true)),
                             StructField("StringCol", StringType)))

val rdd = spark.sparkContext.parallelize(testRows)

//Save DataFrame as TFRecords
val df: DataFrame = spark.createDataFrame(rdd, schema)
df.write.format("tfrecord").option("recordType", "Example").save(path)

//Read TFRecords into DataFrame.
//The DataFrame schema is inferred from the TFRecords if no custom schema is provided.
val importedDf1: DataFrame = spark.read.format("tfrecord").option("recordType", "Example").load(path)
importedDf1.show()

//Read TFRecords into DataFrame using custom schema
val importedDf2: DataFrame = spark.read.format("tfrecord").schema(schema).load(path)
importedDf2.show()

Use partitionBy

The following example shows to how to use partitionBy, which is not supported by Spark Tensorflow Connector

// launch spark-shell with the following command:
// SPARK_HOME/bin/spark-shell --jar target/spark-tfrecord_2.12-0.3.0.jar

import org.apache.spark.sql.SaveMode

val df = Seq((8, "bat"),(8, "abc"), (1, "xyz"), (2, "aaa")).toDF("number", "word")
df.show

// scala> df.show
// +------+----+
// |number|word|
// +------+----+
// |     8| bat|
// |     8| abc|
// |     1| xyz|
// |     2| aaa|
// +------+----+

val tf_output_dir = "/tmp/tfrecord-test"

// dump the tfrecords to files.
df.repartition(3, col("number")).write.mode(SaveMode.Overwrite).partitionBy("number").format("tfrecord").option("recordType", "Example").save(tf_output_dir)

// ls /tmp/tfrecord-test
// _SUCCESS        number=1        number=2        number=8

// read back the tfrecords from files.
val new_df = spark.read.format("tfrecord").option("recordType", "Example").load(tf_output_dir)
new_df.show

// scala> new_df.show
// +----+------+
// |word|number|
// +----+------+
// | bat|     8|
// | abc|     8|
// | xyz|     1|
// | aaa|     2|

Contributing

Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.

License

This project is licensed under the BSD 2-CLAUSE LICENSE - see the LICENSE.md file for details