Nd4 j slides.pptx
Download as PPTX, PDF•4 likes•1,384 views
nd4j is a scientific computing library for Java on the JVM, providing support for ND arrays and corresponding mathematical operations similar to libraries like NumPy and MATLAB. It incorporates various BLAS implementations and allows for matrix manipulation with features such as adaptive learning rates and solvers. The library is designed to be cross-platform, supporting desktop, cloud, and mobile environments, and aims to eliminate fragmentation in scientific computing tools on the JVM.
![Numpy
>>> import numpy as np
>>> a = np.ones((5,5))
>>> np.sum(a,axis=1)
array([ 5., 5., 5., 5., 5.])](https://image.slidesharecdn.com/nd4jslides-150428113103-conversion-gate01/85/Nd4-j-slides-pptx-6-320.jpg)
![Nd4j java/scala
import org.nd4j.api.linalg.DSL._
import org.nd4j.linalg.api.ndarray.INDArray
import org.nd4j.linalg.factory.Nd4j
val a = Nd4j.ones(5,5)
println Nd4j.sum(a,1)
[5.0, 5.0, 5.0, 5.0, 5.0]](https://image.slidesharecdn.com/nd4jslides-150428113103-conversion-gate01/85/Nd4-j-slides-pptx-7-320.jpg)
Nd4 j slides.pptx
- 1. ND4J ND-ARRAYS FOR JAVA SCIENTIFIC COMPUTING ON THE JVM
- 2. Why another scientific library?
- 3. Different libraries: JVM: ● Jblas (not maintained) ● Breeze (scala) ● Spire (scala) ● Core.matrix(clojure) ● Netlib blas ● N + 1 pure java impls Real scientific environments (no fragmentation!): ● octave/matlab ● numpy ● R ● Theano
- 4. Blas and NDArrays BLAS - Basic Linear Algebra Sub programs. Underlies EVERY fast matrix lib out there. NDArrays - matrices, tensors, math is done on axes (eg: rows/cols)
- 5. Matlab/Octave a = ones(5,5) a = 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 octave:2> sum(a,1) ans = 5 5 5 5 5
- 6. Numpy >>> import numpy as np >>> a = np.ones((5,5)) >>> np.sum(a,axis=1) array([ 5., 5., 5., 5., 5.])
- 7. Nd4j java/scala import org.nd4j.api.linalg.DSL._ import org.nd4j.linalg.api.ndarray.INDArray import org.nd4j.linalg.factory.Nd4j val a = Nd4j.ones(5,5) println Nd4j.sum(a,1) [5.0, 5.0, 5.0, 5.0, 5.0]
- 8. Blas Implementations ● Cublas (cuda) ● Openblas ● Intel MKL(cpu) ● Nvblas(cuda) ● Clblas(opencl) ● ATLAS(CPU) ● LAPACK(CPU)
- 9. ND4J A DISTRIBUTED FRAMEWORK FOR MATRIX MANIPULATION NATIVE JBLAS GPU JCUDA ?? WITH SWAPPABLE BACKENDS…
- 10. Core Concepts ● DataBuffer - A DataBuffer is a storage abstraction. A raw buffer could be an nio buffer or a jcuda buffer. This allows for backend optimal storage. ● FFTInstance - An instantiation of FFT. Every hardware maker has their own impls. ● ConvolutionInstance - An instantiation of Convolution. Same as FFT ● Complex Numbers - (An abstraction over different libs) ● BlasWrapper - An interface to a blas backend (blas is a standard for cripe sakes!) ● Op - Some sort of a mathematical operation (dot product, element wise addition,negation,..) ● Garbage Collector (WIP) - Native resource handling where needed
- 11. Other Features ● Loss Functions ● Adaptive learning rates ● Solvers (LBFGS,Conjugate gradient,Hessian Free,SGD) ● RNG: (DIfferent distributions based on apache math)
- 12. ND4J ND4J IS PORTABLE & CROSS-PLATFORM DESKTOP CLOUD MOBILE