michaelnowotny / cocos Goto Github PK

Hiya,

Thanks for writing cocos! It seems to be just what I was looking for, as I want to compute a bunch of large expressions on my GPU!

The documentation for lambdifying was a bit confusing, so I thought I'd ask here. Specifically, the readme.md refers to the LambdifiedMatrixExpression function. It takes the following four arguments:

argument_symbols=argument_symbols,
time_symbol=t,
symbolic_matrix_expression=jacobian_f,
symbolic_time_function_name_to_numeric_time_function_map={'g': numeric_time_function})

I'm happy with the argument_symbols and symbolic_matrix_expression, these seem to be equivalent to the args and expr arguments that sp.lambdify takes. I have no idea what the other two mean. Could you expand on them?

The following code is a slight simplification of my problem. The following generates a summed spectrum of ten Gaussians with parameters stored in 1D arrays. In addition to this, there is the x variable, which is a 1D array of a different length (100 in the example). With reference to the cocos lambdify docs, I could perhaps guess that the gaussian parameters are the "argument symbols", and x is the "time_symbol". But for my real problem, I am computing 2D Gaussians, and have two such variables, x and y, which are themselves 2D arrays containing the coordinates to evaluate the expression at. The time_symbol argument hints at only one variable, not two, so I am a bit confused.

At this point I decided to post here, in case you could shed some light on what I should do.

import sympy as sp
import matplotlib.pyplot as plt
import numpy as np

def Gaussian(x, a, xc, sigma):
    return a*sp.exp(-((x-xc)**2)/(2*sigma**2))

x, i, n = sp.symbols('x i n') # x is our main independent variable. i and n are indices
a, xc, sigma = sp.symbols('A xc sigma', cls=sp.IndexedBase) # parameters that are held in arrays
symbols = (x, a, xc, sigma, n)

Gauss = Gaussian(x, a[i], xc[i], sigma[i])
model = sp.Sum(Gauss, (i, 0, n-1)) # Sum over the parameters from (i=0 to i=n)
func = sp.lambdify(symbols, model, modules = 'numpy')

X = np.arange(100)
N = 10
A, XC, SIGMA = np.random.random((3, N))*100
y = func(X, A, XC, SIGMA, N)

plt.figure()
plt.plot(y)

I guess this is not much of an actual issue (except maybe it could result in an added paragraph in the README!), but I didn't see any better place of asking this. How does cocos relate to the cupy project? Are there any major paradigm or performance differences?

Hi, I'm getting an error when I run the following line in the python interpreter or if it is run within a script:
import cocos.numerics as cn

This returns the following:
Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/cocos/numerics/__init__.py", line 4, in <module> import cocos.numerics.random as random AttributeError: module 'cocos' has no attribute 'numerics'

I'm using a Mac (OS X Catalina 10.15.4)
cd.info() does work:
Cocos running on ArrayFire v3.6.4 (OpenCL 64bit) [0] Apple: Intel(R) Iris(TM) Plus Graphics 640 | OpenCL | compute version 1.2

Thanks
Mike