def make_plot():
x = []
#battle#
y1 = [] #originalが勝つ割合
y2 = [] #svdが勝つ割合
y3 = [] #引き分けの割合
#frobenius#
y4 = []
X = get_np.reshape(27, 27, 27)
norm = np.sqrt(np.sum(X * X))
for i in range(0, 28):
#battle#
y1.append(main.battle(get_np, tucker(get_np, i)[0])[0])
y2.append(main.battle(get_np, tucker(get_np, i)[0])[1])
y3.append(main.battle(get_np, tucker(get_np, i)[0])[2])
#frobenius#
Y = tucker(get_np, i)[0].reshape((27, 27, 27))
rate = tucker(get_np, i)[1]
norm1 = np.sqrt(np.sum((X - Y) * (X - Y)))
x.append(rate)
y4.append(norm1 / norm)
#battle#
plt.xlabel("compression ratio")
plt.plot(x, y1, color='red')
plt.plot(x, y2, color='blue')
plt.plot(x, y3, color='green')
#frobenius#
plt.plot(x, y4, color='black')
plt.show()
def save_file():
with open("task2.dat", "w") as f:
for i in range(0, 28):
#圧縮率
x = tucker(get_np, i)[1]
#battle
y1 = []
y2 = []
y3 = []
for _ in range(5):
y1.append(main.battle(get_np,
tucker(get_np, i)[0])[0]) #originalが勝つ割合
y2.append(main.battle(get_np,
tucker(get_np, i)[0])[1]) #svdが勝つ割合
y3.append(main.battle(get_np,
tucker(get_np, i)[0])[2]) #引き分けの割合
y1_m = np.mean(y1)
y2_m = np.mean(y2)
y3_m = np.mean(y3)
y1_std = np.std(y1)
y2_std = np.std(y2)
y3_std = np.std(y3)
#frobenius
y4 = tucker(get_np, i)[2]
f.write("{} {} {} {} {} {} {} {}\n".format(x, y1_m, y2_m, y3_m, y4,
y1_std, y2_std, y3_std))
def cmpr(rate, num_svd, num_hosvd): #圧縮率、svd特異値数、hosvd特異値数
y1 = main.battle(hosvd(get_np, num_hosvd)[0],
svd(get_np, num_svd)[0])[0] #hosvdの勝率
y2 = main.battle(hosvd(get_np, num_hosvd)[0],
svd(get_np, num_svd)[0])[1] #svdの勝利
y3 = main.battle(hosvd(get_np, num_hosvd)[0],
svd(get_np, num_svd)[0])[2] #引き分け
return [rate, y1, y2, y3]
def save_file():
with open("task1.dat", "w") as f:
for r in range(0, 82):
#圧縮率
x = approx(get_np, r)[1]
#battle
y1 = []
y2 = []
y3 = []
for _ in range(5): #5回分のループ
y1.append(main.battle(get_np, approx(get_np, r)[0])[0]) #originalが勝つ割合
y2.append(main.battle(get_np, approx(get_np, r)[0])[1]) #svdが勝つ割合
y3.append(main.battle(get_np, approx(get_np, r)[0])[2]) #引き分けの割合
y1_m = np.mean(y1)
y2_m = np.mean(y2)
y3_m = np.mean(y3)
y1_std = np.std(y1)
y2_std = np.std(y2)
y3_std = np.std(y3)
#frobenius#
y4 = approx(get_np, r)[2]
f.write("{} {} {} {} {} {} {} {}\n".format(x, y1_m, y2_m, y3_m, y4, y1_std, y2_std, y3_std))
def make_plot():
#battle#
x = []
y1 = [] #originalが勝つ割合
y2 = [] #svdが勝つ割合
y3 = [] #引き分けの割合
#frobenius#
y4 = []
X = get_np.reshape(81,243) #もとの行列をreshape
u, s, v = linalg.svd(X) #svd
norm = np.sqrt(np.sum(X * X)) #sのフロベニウスノルム
for r in range(0, 82):
#battle#
y1.append(main.battle(get_np, approx(get_np, r))[0])
y2.append(main.battle(get_np, approx(get_np, r))[1])
y3.append(main.battle(get_np, approx(get_np, r))[2])
#frobenius#
ur = u[:, :r]
sr = np.diag(np.sqrt(s[:r])) #sの平方根
vr = v[:r, :]
A = ur @ sr
B = sr @ vr
Y = A @ B #近似した行列
norm1 = np.sqrt(np.sum((X-Y) * (X-Y))) #フロベニウスノルム
rate = (A.size+B.size) / X.size
y4.append(norm1 / norm)
x.append(rate) #残した特異値の割合
plt.xlabel("compression ratio")
######plt.ylabel("ratio")
#battle#
plt.plot(x, y1, color = 'red')
plt.plot(x, y2, color = 'blue')
plt.plot(x, y3, color = 'green')
#frobenius#
plt.plot(x, y4, color = 'black')
plt.show()