# This uses different computation methods inside (and outside) iarray and compares performance. from time import time import iarray as ia import numpy as np import numexpr as ne # Define array params dtype = np.float64 shape = [10000, 8000] cshape = [1000, 800] bshape = [100, 100] ia.set_config(chunkshape=cshape, blockshape=bshape) dtshape = ia.DTShape(shape, dtype) # Create initial arrays ia1 = ia.linspace(dtshape, 0, 10) np1 = ia.iarray2numpy(ia1) t0 = time() np2 = np.cos(np1) t1 = time() print("Time for numpy evaluation: %.3f" % (t1 - t0)) t0 = time() np3_ = ne.evaluate("cos(np1)") t1 = time() print("Time for numexpr evaluation: %.3f" % (t1 - t0)) t0 = time() expr = ia.expr_from_string("cos(x)", {"x": ia1}) ia2 = expr.eval() t1 = time() print("Time for iarray evaluation: %.3f (cratio: %.2fx)" % ((t1 - t0), ia2.cratio)) np3 = ia.iarray2numpy(ia2) ia.cmp_arrays(np3, np2, "OK. Results are the same.") t0 = time() ia3 = ia.cos(ia1).eval() t1 = time() print("Time for iarray via lazy evaluation: %.3f (cratio: %.2fx)" % ((t1 - t0), ia3.cratio)) np4 = ia.iarray2numpy(ia3) ia.cmp_arrays(np4, np2, "OK. Results are the same.")