def main():
c = 10
fs = 4
table = {}
table[c - 1] = set(f(c - 1))
table[c - 2] = set(f(c - 2))
table[c - 3] = set(f(c - 3))
while 1:
print(c)
if not c in table:
table[c] = set(f(c))
s1, s2, s3, s4 = table[c], table[c - 1], table[c - 2], table[c - 3]
if len(s1) == fs and len(s2) == fs and len(s3) == fs and len(s4) == fs:
print(c - 3)
break
c += 1
def main():
c = 10
fs = 4
table = {}
table[c-1] = set(f(c-1))
table[c-2] = set(f(c-2))
table[c-3] = set(f(c-3))
while 1:
print(c)
if not c in table:
table[c] = set(f(c))
s1,s2,s3,s4 = table[c],table[c-1],table[c-2],table[c-3]
if len(s1)==fs and len(s2)==fs and len(s3)==fs and len(s4)==fs:
print(c-3)
break
c+=1
def train_linear(train_fea, train_tar, test_fea, test_tar, eta, lamb, maxIter, debug) :
dataNum = train_fea.shape[0]
dimNum = train_fea.shape[1]
beta = np.zeros([dimNum+1, 1])
# beta = np.zeros([dimNum, 1])
# Pad leading 1s to the beginging of training data
train_fea = padOne(train_fea, 1)
test_fea = padOne(test_fea, 1)
converged = False
curIter = 0
while not converged :
curIter = curIter + 1
tmp = f(train_fea, beta) - train_tar
new_beta = beta - 2*eta*np.dot(np.transpose(train_fea), tmp) + lamb * beta
diff = np.linalg.norm(new_beta - beta, 2)
beta = new_beta
if debug == 1 :
print('Iter: ', curIter, '\t change: ', diff)
# evaluate training resutls
if curIter % 1000 == 0 :
train_res = f(train_fea, beta)
test_res = f(test_fea, beta)
# calErr(test_res, test_tar)
print('Iter: ', curIter, '\t Train Err: ', calErr(train_res, train_tar), '\t Test Err: ', calErr(test_res, test_tar))
if curIter > maxIter :
converged = True
# train_res = f(train_fea, beta)
# for ct1 in range(test_fea.shape[0]) :
# print (test_fea[ct1,:])
# test_res = f(test_fea, beta)
# print(calErr(test_res, test_tar))
return beta
def to_imgs(body, wheel1, wheel1_center, wheel1_offset, wheel2, wheel2_center,
wheel2_offset):
iterations = 100
for i in range(iterations):
wheel1_rot = rotateImage(wheel1, wheel1_center, 1000 / iterations * i)
wheel2_rot = rotateImage(wheel2, wheel2_center, 1000 / iterations * i)
wheel1_frame = np.zeros((body.shape[0], body.shape[1], 4),
dtype=np.uint8)
wheel1_frame[wheel1_offset[0]:wheel1_offset[0] + wheel1_rot.shape[0],
wheel1_offset[1]:wheel1_offset[1] +
wheel1_rot.shape[1]] = wheel1_rot
merged = blend_transparent(body, wheel1_frame)
wheel2_frame = np.zeros((body.shape[0], body.shape[1], 4),
dtype=np.uint8)
wheel2_frame[wheel2_offset[0]:wheel2_offset[0] + wheel2_rot.shape[0],
wheel2_offset[1]:wheel2_offset[1] +
wheel2_rot.shape[1]] = wheel2_rot
merged = blend_transparent(merged, wheel2_frame)
shifted = shift_left(merged, 1000 / iterations * i)
cv2.imwrite(f('animation/rot_{}.png').format(i), shifted)
beta = np.reshape(beta, (len(tmp), 1))
tmp = mf.readline().rstrip('\n')
tmp = tmp.split(',')
m = np.asarray(list(map(float, tmp)))
tmp = mf.readline().rstrip('\n')
tmp = tmp.split(',')
s = np.asarray(list(map(float, tmp)))
feature = parseTestData(tstFile)
feature = normalizePara(feature, m, s)
feature = padOne(feature, dim=1)
# feature = normalize(feature, axis=0)
gndHat = f(feature, beta)
# for ct1 in range(gndHat.shape[0]) :
# data = np.rint(gndHat[ct1])
# print(int(np.asscalar(data)))
submit = open(outFile, 'w')
header = ['id', 'label']
submit.write(','.join(header) + '\n')
for ct1 in range(gndHat.shape[0]):
data = np.rint(gndHat[ct1])
data = int(np.asscalar(data))
cont = [str(ct1 + 1), str(data)]
submit.write(','.join(cont) + '\n')
""" Main iteration of the algorithm """
f = f.himmelblau
swarm = Swarm(f, param.NPARTICLE)
swarm.initialize()
g = getGlobalBest(swarm)
avg = np.zeros(param.NITERATION)
std = np.zeros(param.NITERATION)
bfit = np.zeros(param.NITERATION)
# plot the level curves
r = np.arange(param.RANGE[0] - .2, param.RANGE[1] + .2, 0.05)
x, y = np.meshgrid(r, r)
z = copy.deepcopy(x) # temporarily
for i in range(x.shape[0]):
z[i, :] = [f([x[i, j], y[i, j]]) for j in range(x.shape[1])]
fig, ax = plt.subplots()
ax.contour(x, y, z, 60)
plt.xlabel("Variable X")
plt.ylabel("Variable Y")
# plot the swarm and the global best particle
pos = np.matrix([[swarm.particles[j].x[0], swarm.particles[j].x[1]]
for j in range(param.NPARTICLE)])
h = ax.plot(pos[:, 0], pos[:, 1], 'ob') # swarm
hg = ax.plot(g.x[0], g.x[1], 'xr', markersize=8) # global best
for i in range(param.NITERATION):
plt.title("Iteration: " + str(i + 1))
for j in range(param.NPARTICLE):
"""
第13章 模块
"""
"""
模块就是程序
封装的几个级别:
容器(list等, 是对数据的封装)->函数(对语句的封装)->类(对属性和方法的封装)->模块
保存的每一个.py都是一个独立的模块
模块的作用: 1. 组织代码更有条理; 2. 代码重用
"""
# 导入方法1, 不推荐
import util # 同一目录下可直接导入, 不需要当前目录是个package(即没__init__)也行
print(util.x)
util.f()
o = util.C()
o.f()
# 起个别名, 强推
import util as ut
print(ut.x)
ut.f()
o = ut.C()
o.f()
# 容易和当前文件命名冲突, 不建议
from util import C
o = C()