Numpy4J
Numpy4J copied to clipboard
JosephCottam
Interface Java with Numpy. Try to be reasonable.
leveraging NumPy from Java makes a lot of sense, but it seems that this project lost steam ? Why? Has something else made it irrelevant ? thanks -Charles
Use the Numpy C bindings, not the python interpreter, to interface with numpy.
Give a chance for zero-configuration development by providing a "default.properties" file that derives java.path, python.path and python.version information from environment variables. The build file loads the system.properties first, then default.properties,...
Testing fails if JUnit is not in the ./lib directory. Either fetch it (ivy??) or get the junit install location from the system.
On the java side, implement the indexer so you can array[1,2...] On the python side, unpack the information (probably in NPType) so the dimensionality is preserved.
On the java side, implement the accessors On the bridge, translate the striding information
NPType dictates the DType and other flags. It is not currently unpacked. Probably unpack following the pattern established in the XXXX_py2J methods (except it would be XXXX J2py). Using non-default...
Allow a target NPArray to be passed through the JNI bridge so results arrays can be pre-allocated (possibly recycled) on the Java side.
Build up a worklist on the java side to be executed entirely on the python side without recrossing the JNI boundary. May be similar to RDD used in Spark.
Metadata
Owner
Metadata
Interface Java with Numpy. Try to be reasonable.