faydynet packages DyNet for direct integration into Scala/Java environments on Linux, Mac, and Windows. No other downloads or development tools are necessary other than Java and sbt.
fatdynet is being used by at least two other clulab projects, and sometimes the best way to understand something is to see an example. Here are some links for those who want to cut to the chase:
- https://github.com/clulab/processors/blob/master/main/build.sbt#L21
- https://github.com/clulab/processors/blob/master/main/src/main/scala/org/clulab/dynet/
- https://github.com/clulab/factuality/blob/master/build.sbt#L16
- https://github.com/clulab/factuality/blob/master/src/main/scala/org/clulab/factuality/Factuality.scala
To use fatdynet with sbt in your own project, first check the releases page for the most recent version number and then make a build.sbt file like
name := "fatdynetClient"
organization := "org.clulab"
scalaVersion := "2.12.4"
libraryDependencies ++= Seq(
"org.clulab" %% "fatdynet" % "0.2.4"
)
It can be tested with a program like
package org.clulab.fatdynetClient.apps
import edu.cmu.dynet.examples.XorScala
object XorScalaClientApp extends App {
XorScala.main(args)
}
A clone of this project should work out of the box with sbt. You can perform a quick check with sbt test
which should pass or sbt run which should produce output similar to the lines quoted below.
> git clone https://github.com/clulab/fatdynet
> cd fatdynet
> sbt
sbt: fatdynet> test
...
sbt: fatdynet> run
...
[info] Running org.clulab.fatdynet.apps.XorScalaApp
Running XOR example
[dynet] random seed: 2354967643
[dynet] allocating memory: 512MB
[dynet] memory allocation done.
Dynet initialized!
Computation graphviz structure:
digraph G {
rankdir=LR;
nodesep=.05;
N0 [label="v0 = parameters({8,2}) @ 0x7fe79abb0478"];
N1 [label="v1 = parameters({8}) @ 0x7fe79abb05a8"];
N2 [label="v2 = parameters({1,8}) @ 0x7fe79abb06f8"];
N3 [label="v3 = parameters({1}) @ 0x7fe79abb08a8"];
N4 [label="v4 = constant({2})"];
N5 [label="v5 = scalar_constant(0x7fe79abb9060)"];
N6 [label="v6 = v0 * v4"];
N0 -> N6;
N4 -> N6;
N7 [label="v7 = v6 + v1"];
N6 -> N7;
N1 -> N7;
N8 [label="v8 = tanh(v7)"];
N7 -> N8;
N9 [label="v9 = v2 * v8"];
N2 -> N9;
N8 -> N9;
N10 [label="v10 = v9 + v3"];
N9 -> N10;
N3 -> N10;
N11 [label="v11 = || v10 - v5 ||^2"];
N10 -> N11;
N5 -> N11;
}
Training...
iter = 0, loss = 2.2457085
iter = 1, loss = 1.6314303
iter = 2, loss = 1.2347934
iter = 3, loss = 0.91187227
iter = 4, loss = 0.63016045
iter = 5, loss = 0.3049145
iter = 6, loss = 0.09319025
iter = 7, loss = 0.019082734
iter = 8, loss = 0.003395962
iter = 9, loss = 5.326783E-4
iter = 10, loss = 8.053323E-5
iter = 11, loss = 1.1818531E-5
iter = 12, loss = 1.7824146E-6
iter = 13, loss = 2.6764963E-7
iter = 14, loss = 4.260896E-8
iter = 15, loss = 6.514804E-9
iter = 16, loss = 1.1485586E-9
iter = 17, loss = 1.756435E-10
iter = 18, loss = 3.5814907E-11
iter = 19, loss = 4.943601E-12
iter = 20, loss = 1.2052581E-12
iter = 21, loss = 6.865619E-13
iter = 22, loss = 3.5704772E-13
iter = 23, loss = 2.1049829E-13
iter = 24, loss = 1.687539E-13
iter = 25, loss = 6.57252E-14
iter = 26, loss = 2.1316282E-14
iter = 27, loss = 7.1054274E-15
iter = 28, loss = 1.0658141E-14
iter = 29, loss = 7.1054274E-15
[success] Total time: 1 s, completed Sep 17, 2018 5:20:44 PM
The source code used for the libraries of this project is based on
- dynet commit d9d3e80 (version 2.1)
- eigen commit b2e267d
with some modifications explained on the project's wiki pages.
The build environment included
- java 1.8.0_162
- sbt 1.2.1
The files in lib-2.12 were built with
- scala 2.12.4
cmake .. -DSCALA_VERSION=2.12.4 -DEIGEN3_INCLUDE_DIR=../../eigen -DENABLE_CPP_EXAMPLES=ON -DENABLE_SWIG=ONmake
and those in lib-2.11 were built with
- scala 2.11.11
cmake .. -DSCALA_VERSION=2.11.11 -DEIGEN3_INCLUDE_DIR=../../eigen -DENABLE_CPP_EXAMPLES=ON -DENABLE_SWIG=ONmake
For more details see Compiling DyNet for Scala.