Collecting results from a loop returning NumPy arrays

Question

I am admittedly a fairly simple Python programmer trying to learn when I encounter problems related to various research issues. And I hit one of these problems - in particular, how to handle loops, where I return a bunch of data, rather than the usual "exit from one number" examples, where you simply add the loop result to the previous one.

There is a Gist unlooped script here I'm trying to run: https://gist.github.com/1390355

In fact, the important point is the end of the model_solve function:

def model_solve(t): # lots of variables set params = np.zeroes((n_steps,n_params) params[:,0] = beta params[:,1] = gamma timer = np.arange(n_steps).reshape(n_steps,1) SIR = spi.odeint(eq_system, startPop, t_interval) output = np.hstack((timer,SIR,params)) return output 

This returns the results of the ODE integration bit ( spi.odeint ) along with the simple “What step are we now?” a timer and essentially two columns, the values ​​of two random variables are repeated many, many times in the form of an array of 4950 rows and 7 NumPy columns.

However, the goal is to run a Monte Carlo analysis of two parameters (beta and gamma) that have random values. Essentially, I want to create a function that will alternate like this:

 def loop_function(runs): for i in range(runs): model_solve(100) # output of those model_solves collected here # return collected output 

Then the collected output will be written to a file. Normally, I would simply use each model_solve function for its own results for the file, but this code will be run on PiCloud or another platform where I do not have the ability to write the file until the results are returned to the local computer. Instead, I am trying to get a huge NumPy array from the columns runs * 7 and 4950 rows, which can then be written to a file on my local machine.

Any tips on how to approach this?

Answer 1

use list to save all results:

 results = [] for i in range(runs): results.append(model_solve(100)) 

then we get the output array:

 np.hstack(results) 

Answer 2

In fact, if your code has a much larger loop, you should always try to vectorize your problem. If speed is important, you should know that “for loops” is the neck of the bottle. In addition, the add operation is very slow and requires more memory because it creates copies. So, the best solution should be:

 results = [0]*runs # if you want to use lists... [model_solve(100) for x in results] # do see list comprehension in python 

Besides using a list, you can also use an array to store the results:

 resutls=np.zeros([numberOfRuns,ShapeOfModelResults]) for i in range(numberOfRuns): results[numberOfRuns,modelSolve(100)] # this will put the result directly in the matrix 

I hope my answer helps you write faster and more understandable code.