diff options
Diffstat (limited to 'buch/papers/multiplikation/code/MM.py')
| -rw-r--r-- | buch/papers/multiplikation/code/MM.py | 90 |
1 files changed, 59 insertions, 31 deletions
diff --git a/buch/papers/multiplikation/code/MM.py b/buch/papers/multiplikation/code/MM.py index 626b82d..8057850 100644 --- a/buch/papers/multiplikation/code/MM.py +++ b/buch/papers/multiplikation/code/MM.py @@ -5,6 +5,7 @@ Created on Fri Mar 19 07:31:29 2021 @author: nunigan """ +import scipy.stats import numpy as np import time import matplotlib.pyplot as plt @@ -132,6 +133,7 @@ def winograd2(A, B): return C def test_perfomance(n): + t_mm = [] t_mm_dc = [] t_mm_strassen = [] @@ -174,17 +176,18 @@ def test_perfomance(n): plt.plot(n, t_mm_strassen, label='Strassen', lw=5) plt.plot(n, t_wino, label='Winograd', lw=5) plt.plot(n, t_np, label='NumPy A@B', lw=5) + # plt.xscale('log', base=2) plt.legend() plt.xlabel("n") plt.ylabel("time (s)") - plt.grid(True) + plt.grid(True, which="both", ls="-") plt.tight_layout() # plt.yscale('log') plt.legend(fontsize=19) plt.savefig('meas_' + str(max(n))+ '.pdf') arr = np.array([n, t_mm, t_mm_dc, t_mm_strassen, t_wino, t_np]) np.savetxt('meas_' + str(max(n))+ '.txt',arr) - return arr + return t_np def plot(num): @@ -198,10 +201,11 @@ def plot(num): plt.plot(n, t_mm, label='3 For Loops', lw=5) plt.plot(n, t_mm_dc, label='Divide and Conquer', lw=5) plt.plot(n, t_mm_strassen, label='Strassen', lw=5) - # plt.plot(n, t_wino, label='Winograd', lw=5) + plt.plot(n, t_wino, label='Winograd', lw=5) plt.plot(n, t_np, label='NumPy A@B', lw=5) plt.legend() plt.xlabel("n") + # plt.yscale('log', base=10) plt.ylabel("time (s)") plt.grid(True) plt.tight_layout() @@ -211,36 +215,39 @@ def plot(num): return arr def plot_c_res(ave, num): + MM = np.loadtxt("meas/MM.txt", delimiter=',') - # winograd = np.loadtxt("meas/winograd.txt", delimiter=',') + winograd = np.loadtxt("meas/winograd.txt", delimiter=',') blas = np.loadtxt("meas/blas.txt", delimiter=',') MM_dc = np.loadtxt("meas/MM_dc.txt", delimiter=',') strassen = np.loadtxt("meas/strassen.txt", delimiter=',') MM_t = MM[:,0] MM_n = MM[:,1] - MM_t = np.mean(MM_t.reshape(-1,ave),axis=1) - MM_n = np.mean(MM_n.reshape(-1,ave),axis=1) + # MM_t = np.mean(MM_t.reshape(-1,ave),axis=1) + # MM_n = np.mean(MM_n.reshape(-1,ave),axis=1) MM_dc_t = MM_dc[:,0] MM_dc_n = MM_dc[:,1] - MM_dc_t = np.mean(MM_dc_t.reshape(-1,ave),axis=1) - MM_dc_n = np.mean(MM_dc_n.reshape(-1,ave),axis=1) + # MM_dc_t = np.mean(MM_dc_t.reshape(-1,ave),axis=1) + # MM_dc_n = np.mean(MM_dc_n.reshape(-1,ave),axis=1) strassen_t = strassen[:,0] strassen_n = strassen[:,1] - strassen_t = np.mean(strassen_t.reshape(-1,ave),axis=1) - strassen_n = np.mean(strassen_n.reshape(-1,ave),axis=1) + # strassen_t = np.mean(strassen_t.reshape(-1,ave),axis=1) + # strassen_n = np.mean(strassen_n.reshape(-1,ave),axis=1) - # winograd_t = winograd[:,0] - # winograd_n = winograd[:,1] + winograd_t = winograd[:,0] + winograd_n = winograd[:,1] # winograd_t = np.mean(winograd_t.reshape(-1,ave),axis=1) # winograd_n = np.mean(winograd_n.reshape(-1,ave),axis=1) blas_t = blas[:,0] blas_n = blas[:,1] - blas_t = np.mean(blas_t.reshape(-1,ave),axis=1) - blas_n = np.mean(blas_n.reshape(-1,ave),axis=1) + # blas_t = np.mean(blas_t.reshape(-1,ave),axis=1) + # blas_n = np.mean(blas_n.reshape(-1,ave),axis=1) + + def func(x, a,b): return b*x**a @@ -254,14 +261,16 @@ def plot_c_res(ave, num): plt.rc('axes', labelsize=23) plt.rc('xtick', labelsize=23) plt.rc('ytick', labelsize=23) - plt.plot(MM_n, MM_t, label='3 For Loops', lw=5) - # plt.plot(winograd_n, winograd_t, label='Winograd MM', lw=5) - plt.plot(blas_n, blas_t, label='Blas', lw=5) - plt.plot(strassen_n, strassen_t, label='Strassen', lw=5) - plt.plot(MM_dc_n, MM_dc_t, label='Divide and Conquer', lw=5) + plt.loglog(MM_n, MM_t, '.', label='3 For Loops', lw=5) + plt.loglog(winograd_n, winograd_t, '.', label='Winograd MM', lw=5) + plt.loglog(blas_n, blas_t, '.', label='Blas', lw=5) + plt.loglog(strassen_n, strassen_t, '.', label='Strassen', lw=5) + plt.loglog(MM_dc_n, MM_dc_t, '.', label='Divide and Conquer', lw=5) plt.xlabel("n") + # plt.yscale('log', base=10) + # plt.xscale('log', base=2) plt.ylabel("time (s)") - plt.grid(True) + plt.grid(True, which="both", ls="-") plt.tight_layout() plt.legend(fontsize=19) plt.savefig('c_meas_' + str(num)+ '.pdf') @@ -271,23 +280,42 @@ def plot_c_res(ave, num): # plt.plot(blas_n, func(blas_n, *popt2), 'r-', label='fit MM: a=%5.5f, b=%5.10f' % tuple(popt2)) plt.legend() + # return [MM_n,winograd_n,blas_n,strassen_n,MM_dc_n] + + return [MM_t,winograd_t,blas_t,strassen_t,MM_dc_t] + + +def mean_confidence_interval(data, confidence=0.95): + a = 1.0 * np.array(data) + n = len(a) + m, se = np.mean(a), scipy.stats.sem(a) + h = se * scipy.stats.t.ppf((1 + confidence) / 2., n-1) + return m, h # test%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if __name__ == '__main__': - plot_c_res(1, 4096) - - - # plot(8) - # n = np.logspace(1,10,10,base=2,dtype=(np.int)) + # A = plot_c_res(10, 4096) + # name = ['MM', 'Wino', 'blas', 'strassen', 'dc'] + # for i in range(5): + # ci_inner = [] + # print(name[i]) + # for j in range(11): + # m,h=mean_confidence_interval(A[i][j*10:(j+1)*10]) + # print("({},{})".format(2**(j+1),m)) + # np.savetxt('meas/ci/' + name[i]+'.txt',ci_inner) + + arr = plot(4096) + # n = np.logspace(1,12,12,base=2,dtype=(np.int)) + # n=[2048,4096] # n = np.arange(1,50,2) - A = np.random.randint(-10, 10, (5,3)) - B = np.random.randint(-10, 10, (3,5)) + # A = np.random.randint(-10, 6, (5,3)) + # B = np.random.randint(-10, 6, (3,5)) - C = winograd2(A, B) - C_test = A@B - print(C) - print(C_test) + # C = winograd2(A, B) + # C_test = A@B + # print(C) + # print(C_test) # print(np.equal(C, C_test)) # t_np = test_perfomance(n) |