RavOp

RavOP

Github

Ravop is the Raven Distribution framework’s (RDF) operation library. Ops are a fundamental unit of RDF. Ravop Library contains the main object data types Tensor and Scalar upon which we can perform the various arithmetic and mathematical operations available in the ravop library in a distributed and decentralized manner.

###Ravop in RDF Ravop is responsible for creation of ops and retrieving ops from ravcom and ravsock.

Setup and Installation

Create a virtual environment

virtualenv ravop -p python3

Activate the virtual environment

source ravop/bin/activate

Install RavOp

pip install https://github.com/ravenprotocol/ravop.git

This step will automatically install the dependencies

Set RAVSOCK_SERVER_URL in the ravop/config.py file

RAVSOCK_SERVER_URL = "http://0.0.0.0:9999/"

Note: Make sure ravsock is up and running

Initializing Tensors , Scalars and Graphs

Tensors:Raven Tensors are data type objects which are multidimensional arrays

Scalars:Raven data type object for int or float type.

Graph: For evaluating all the ops after creating it in a computational graph.

These data types can be initialized as follows:

import ravop.core as R

a=R.Tensor([1,2,3])

b=R.Scalar(10)

Example

We can use raven ops to operate on these Tensors. Let's consider adding two Tensors . Adding two tensors can be achieved using the R.add() operation

import ravop.core as R
#inittializing Tensors
a=R.Tensor([1,2,3])
b=R.Tensor([2,3,4])

#using add operation from ravop.core
c=R.add(a,b)

Raven supports Unary and Binary ops :

unary ops

import ravop.core as R
#inittializing Tensor
a=R.Tensor([1,2,3])
#using min operation from ravop.core
c=R.min(a)

Binary ops :

import ravop.core as R
#inittializing Tensors
a=R.Tensor([1,2,3])
b=R.Tensor([2,3,4])
#using multiply operation from ravop.core
c=R.multiply(a,b)

##Supported Ops Arithmetic Ops

Op name	parameters
lin
add
mul
sub
pos
neg
exp
natlog
pow
square
cube
square_root
cube_root
abs

Comparision Opa

Op name	parameters
greater
greater_equal
less
less_equal
equal
not_equal

Logical Operations

Op name	parameters
logical_and
logical_or
logical_not
logical_xor

Statistical Ops

Op name	parameters
mean
average
mode
median
variance
std
percentile

Tensor Ops

Op name	parameters
random
bincount
where
sign
foreach
one_hot_encoding
matmul
multiply
dot
transpose
sum
sort
split
reshape
concat
min
max
unique
argmax
argmin
expand_dims
inv
gather
reverse
stack
tile
slice
find_indices
shape

Federated Learning

Op name	parameters
federated_mean
federated_variance
federated_standard_deviation