def __calc__(self):
p = Factorial
iter = self.__state_iter
proxy = SolverProxy()
num = 0
for tau, tau_state in iter(p.tau_lh()):
for delta, delta_state in iter(p.delta_lh()):
for cn, cn_state in iter(p.cn_lh()):
for cr, cr_state in iter(p.cr_lh(cn, delta)):
for s, s_state in iter(p.s_lh(cn)):
for a, a_state in iter(p.a_lh()):
# calculate model with restocking_fee
par_nb = Parameter(MODEL_NB, tau=tau, a=a, s=s, cn=cn)
sol_nb = proxy.read_or_calc_and_write(par_nb, resolution='middle')
num += 1; print(num)
#sol_nb = ModelNBSolver.solve(par_nb, 'low')
key_n = self.__key(MODEL_NB, tau_state, a_state, s_state, cr_state, cn_state, delta_state)
self._add(key_n, par_nb, sol_nb)
# calculate model two
par_o = Parameter(MODEL_2, tau=tau, a=a, s=s, cr=cr, cn=cn, delta=delta)
sol_o = proxy.calculate(par_o)
key_o = self.__key(MODEL_2, tau_state, a_state, s_state, cr_state, cn_state, delta_state)
self._add(key_o, par_o, sol_o)
proxy.commit()
def _a_cn_generator(self):
"""
Helper generator to generate parameters for plotting
it varies cn from [cr, 1] and a from [0, upper_bound_a]
Returns a tuple of (par_model_1, par_model_2)
"""
def __cr(cn):
#if self.cr == 'delta*cn/2':
# return self.delta * cn / 2
if self.cr == '0.4*cn':
return 0.4 * cn
else:
return self.cr
def __s(cn):
if self.s == 'cn/2':
return cn / 2
else:
return self.s
if self.s == '0.4*cn' and self.cr == '0.4*cn':
for line, cn in enumerate(
drange(self.lower_bound_cn, self.upper_bound_cn,
self.step_size_cn)):
for col, a in enumerate(
drange(self.lower_bound_a, self.upper_bound_a,
self.step_size_a)):
par_model_1 = Parameter(MODEL_NB,
tau=self.tau,
a=a,
s=0.4 * cn,
cn=cn)
par_model_2 = Parameter(MODEL_2_QUAD,
tau=self.tau,
a=a,
s=0.4 * cn,
cr=0.4 * cn,
cn=cn,
delta=self.delta)
yield (line, col, par_model_1, par_model_2)
else:
for line, cn in enumerate(
drange(self.lower_bound_cn, self.upper_bound_cn,
self.step_size_cn)):
for col, a in enumerate(
drange(self.lower_bound_a, self.upper_bound_a,
self.step_size_a)):
par_model_1 = Parameter(MODEL_NB,
tau=self.tau,
a=a,
s=__s(cn),
cn=cn)
par_model_2 = Parameter(MODEL_2_QUAD,
tau=self.tau,
a=a,
s=__s(cn),
cr=__cr(cn),
cn=cn,
delta=self.delta)
yield (line, col, par_model_1, par_model_2)
def test_something(self):
proxy = SolverProxy()
proxy.read_or_calc_and_write()
tau = 0.35
delta = 0.85
a = 0.001
s = 0
cr = 0.2125
cn = 0.5
par_nb = Parameter(MODEL_NB, tau=tau, a=a, s=s, cn=cn)
sol_nb = ModelNBSolver.solve(par_nb, 'very high')
print('wn = {}, b={}, rho={}, pn={}'.format(sol_nb.dec.wn,
sol_nb.dec.b,
sol_nb.dec.rho,
sol_nb.dec.pn))
print(sol_nb.profit_man)
print(sol_nb.profit_ret)
par_o = Parameter(MODEL_2,
tau=tau,
a=a,
s=s,
cr=cr,
cn=cn,
delta=delta)
sol_o = proxy.calculate(par_o)
print(sol_o.profit_man)
def __calc__(self):
p = Factorial
iter = self.__state_iter
proxy = SolverProxy()
num = 0
for tau, tau_state in iter([p.tau_lo(), p.tau_hi()]):
for cn, cn_state in iter([p.cn_lo(), p.cn_hi()]):
for s, s_state in iter([p.s_lo(), p.s_hi(cn)]):
for a, a_state in iter([p.a_lo(), p.a_hi()]):
# calculate model with restocking_fee
par_nb = Parameter(MODEL_NB, tau=tau, a=a, s=s, cn=cn)
sol_nb = proxy.read_or_calc_and_write(
par_nb, resolution='high')
num += 1
print(num)
#sol_nb = ModelNBSolver.solve(par_nb, 'low')
key_nb = self.__key(MODEL_NB, tau_state, a_state,
s_state, Factorial.HIGH, cn_state,
Factorial.HIGH)
self._add(key_nb, par_nb, sol_nb)
# calculate model without online store
par_no = Parameter(MODEL_1, tau=tau, a=a, s=s, cn=cn)
sol_no = proxy.calculate(par_no)
key_no = self.__key(MODEL_1, tau_state, a_state,
s_state, Factorial.HIGH, cn_state,
Factorial.HIGH)
self._add(key_no, par_no, sol_no)
proxy.commit()
def __both_solutions(self, tau, a, s, cr, cn, delta):
par_n = Parameter(MODEL_1, tau=tau, a=a, s=s, cn=cn)
sol_n = self.__proxy.calculate(par_n)
par_o = Parameter(MODEL_2,
tau=tau,
a=a,
s=s,
cr=cr,
cn=cn,
delta=delta)
sol_o = self.__proxy.calculate(par_o)
return (sol_n, sol_o)
def test_db_and_plot():
from matplotlib import pyplot as plt
from solver import SolverProxy
proxy = SolverProxy()
a = [float(a) for a in np.linspace(0.0, 0.01, num=48 + 2)]
x = np.zeros((len(a), 1))
y = np.zeros((len(a), 1))
for i, act_a in enumerate(a):
if act_a == 0:
x[0] = None
y[0] = None
continue
par = Parameter(MODEL_2_QUAD, tau=0.09, a=act_a, s=0.04000000000000001, cn=0.1, cr=0.04000000000000001, delta=0.7956)
#solP = ModelTwoQuadSolver.solve(par)
solP = proxy.calculate(par, resolution='very-high')
x[i] = solP.dec.wn
y[i] = solP.profit_ret
# plot actual a on x[i]
#plt.text(act_a, x[i], str(act_a))
proxy.commit()
plt.plot(a, x, label='x', marker='o')
plt.plot(a, y, label='y')
plt.legend()
#plt.savefig('figure_quad.pdf')
plt.show()
def plot_model_n_retailer(wn):
par = Parameter(MODEL_1, tau=0.15, cn=0.1, s=0.5 * 0.1, a=0.005)
NUM_PN = 10
NUM_RHO = 100
values = np.zeros([int(NUM_PN * NUM_RHO), 3])
i = 0
for pn in np.linspace(0, 1, NUM_PN):
for rho in np.linspace(1, 20, NUM_RHO):
profit = ModelOneNumericalSolver.get_retailer_profit(
par, wn, pn, rho)
values[i, :] = [pn, rho, profit]
i += 1
# plot
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.plot_trisurf(values[:, 0], values[:, 1], values[:, 2])
ax.set_xlabel('pn')
ax.set_ylabel('rho')
ax.set_title('Retailer\'s Profit, $w_n$={}'.format(wn), loc='left')
# plot optimum
best_pn, best_rho = ModelOneNumericalSolver.get_retailer_decision(par, wn)
best_profit = ModelOneNumericalSolver.get_retailer_profit(
par, wn, best_pn, best_rho)
latexlegend = r'${p_n}^*=$' + '{}'.format(
best_pn) + r', ${\rho}^*=$' + '{:.2f}'.format(best_rho)
ax.scatter(best_pn,
best_rho,
best_profit,
values[:, 2],
color='red',
label=latexlegend)
plt.legend()
plt.show()
def test_optimize_instance_c(self):
solver = ModelTwoNumericalSolver()
par = Parameter(MODEL_2, tau=.0, a=.01, s=.0, cr=.0, cn=.1, delta=.8)
sol = solver.optimize(par)
self.assertIsNotNone(sol)
self.assertAlmostEqual(sol.profit_man, 0.1687500000000)
self.assertAlmostEqual(sol.profit_ret, 0.0281250000000)
def plottiplotti():
proxy = SolverProxy()
a = [float(a) for a in np.linspace(0.0, 0.01, num=48 + 2)]
x = np.zeros((len(a), 1))
y = np.zeros((len(a), 1))
dist = 0
for i, act_a in enumerate(a):
if act_a == 0:
x[0] = None
y[0] = None
continue
par = Parameter(MODEL_2,
tau=0.09,
a=act_a,
s=0.04000000000000001,
cn=0.1,
cr=0.04000000000000001,
delta=0.7956)
solP = ModelTwoSolver.solve(par)
solA = proxy.calculate(par)
x[i] = solP.dec.wn
y[i] = solA.dec.wn
dist += abs(solP.dec.wn - solA.dec.wn)
from matplotlib import pyplot as plt
plt.plot(a, x, label='search')
plt.plot(a, y, label='analy')
plt.legend()
plt.show()
print(dist)
def test_optimize_instance_d(self):
solver = ModelTwoNumericalSolver()
# i dont know whether this parms lead to two a, but i will check the output anyway
par = Parameter(MODEL_2, tau=.4, a=.01, s=.0, cr=.0, cn=.0, delta=.9)
sol = solver.optimize(par)
self.assertIsNotNone(sol)
self.assertAlmostEqual(sol.profit_man, 0.1669565217391)
def test_write_and_read(self):
proxy = SolverProxy()
#par = Parameter(MODEL_2_QUAD, tau=.09, a=0.00146, s=.04, cr=.04, cn=.1, delta=.7956)
par = Parameter(MODEL_1, tau=0.09, a=0.00146, s=.04, cn=.1)
sol = proxy.read_or_calc_and_write(par, comment='unittest')
proxy.commit()
print(sol)
def plot_model_n_manufacturer_both_cases():
par = Parameter(MODEL_1, tau=0.15, cn=0.1, s=0.5 * 0.1, a=0.005)
# print whole profit function of the manufacturer
NUM_POINTS = 100
all_wn = np.linspace(0, 1, NUM_POINTS)
all_profit_one = np.zeros(NUM_POINTS)
all_profit_two = np.zeros(NUM_POINTS)
all_profit = np.zeros(NUM_POINTS)
knee_point = -1
for i, wn in enumerate(all_wn):
all_profit_one[i] = ModelOneNumericalSolver.get_manufacturer_profit(
par, wn, case=_CASE_ONE)
all_profit_two[i] = ModelOneNumericalSolver.get_manufacturer_profit(
par, wn, case=_CASE_TWO)
all_profit[i] = ModelOneNumericalSolver.get_manufacturer_profit(
par, wn)
if all_profit_two[i] == all_profit[i] and knee_point < 0:
knee_point = i
plt.plot(all_wn, all_profit)
plt.text(0.35, 0.085, r'$\rho > 1$', size=8, color='#1f77b4', rotation=38)
plt.text(0.85, 0.048, r'$\rho = 1$', size=8, color='#1f77b4', rotation=-54)
plt.plot(all_wn[10:knee_point],
all_profit_two[10:knee_point],
linestyle='--',
color='grey')
plt.text(0.30, 0.059, r'$\rho = 1$', size=8, color='grey', rotation=38)
plt.title(
r"Manufacturer's Profit $\tau=0.15, c_n =0.1, s=\frac{c_n}{2}, a=0.005$"
)
plt.xlim(0, 1)
plt.ylim(np.nanmin(all_profit) * 1.1, np.nanmax(all_profit) * 1.1)
plt.xlabel('$w_n$')
plt.show()
def test_zwei(self):
from solver import ModelTwoSolver, ModelNBSolver
import numpy as np
np.seterr(all='ignore')
for a in np.linspace(0.024329896907216497, 0.024742268041237116, 10):
par_nb = Parameter(MODEL_NB, tau=0.15, cn=0.1, s=0.5 * 0.1, a=a)
sol = ModelNBSolver.solve(par_nb, resolution='case study')
print(sol.profit_man, sol.profit_ret, sol.dec.wn, sol.dec.b,
sol.case, sol.dec.rho)
def test_something(self):
from solver import ModelTwoSolver
par = Parameter(MODEL_2_QUAD,
tau=0.09,
a=0.0008163265306122449,
s=0.04000000000000001,
cn=0.1,
cr=0.04000000000000001,
delta=0.7956)
ModelTwoSolver.solve(par)
def test_ob_gleich():
par_21 = Parameter(MODEL_2_QUAD, tau=0.09, a=0.004081632653061225, s=0.04000000000000001, cn=0.1, cr=0.04000000000000001, delta=0.7956)
wn_pr_21 = (0.5429821819318537, 0.48640200062519534)
wn_pr_20 = (0.5461081587996248, 0.4885901844326352)
# now test with wn,pr combination of par
sol_with_21 = ModelTwoQuadSolver.getSolution(par_21, wn_pr_21[0], wn_pr_21[1])
sol_with_20 = ModelTwoQuadSolver.getSolution(par_21, wn_pr_20[0], wn_pr_20[0])
assert sol_with_21.profit_man == 0.15756762319737758
print(sol_with_21.profit_man)
print(sol_with_20.profit_man)
def case_quad_fixed_a():
cn = np.linspace(0.0, 0.9, num=100)
y = []
for act_cn in np.linspace(0.0, 0.9, num=100):
act_cn = float(act_cn)
par = Parameter(MODEL_2_QUAD, tau=0.09, a=.4, s=.1, cr=.2, cn=act_cn, delta=.4)
sol = ModelTwoQuadSolver.solve(par, 'low')
y.append(sol.dec.qr)
plt.plot(cn, y)
print(y)
plt.show()
def main():
par = Parameter(MODEL_2_QUAD, tau=0.09, a=0.0038775510204081634, s=0.04000000000000001, cn=0.1, cr=0.04000000000000001, delta=0.7956)
# find any start vector of case 1
startP = ModelTwoQuadGridSearch.search(par, iter=1, case=1)
start_vec = [startP.dec.wn, startP.dec.pr]
# improve wn, pr
result = minimize(ModelTwoQuadSolver._minimize_func, args={'par': par, 'case': 1}, x0=start_vec, method='Nelder-Mead', options={'xatol': 0.000000001, 'fatol': 0.000000001})
# build a new solution using wn, pr
solP = ModelTwoQuadSolver.getSolution(par, float(result.x[0]), float(result.x[1]))
print(startP.profit_man, startP.dec.wn, startP.dec.pr, startP.dec.rho)
print(solP.profit_man, solP.dec.wn, solP.dec.pr, solP.dec.rho)
def __model_one_par_generator(self):
""" Helper generator for yielding all combinations of input data for model 2"""
for tau in drange(0, 1, 0.1):
for a in drange(0.01, 0.1, 0.01):
for s in drange(0, 1, .1):
for cn in drange(s, 1, .1):
yield Parameter(MODEL_1,
tau=round(tau, 1),
a=round(a, 2),
s=round(s, 1),
cn=round(cn, 1))
def test_case_1_or_2(self):
solver = ModelOneNumericalSolver()
par = Parameter(MODEL_1, tau=0.3, a=0.01, s=0, cn=0.3)
# self checking if input vars not in case 1 and not in case 2:
self.assertTrue(
self.__input_is_in_case_1(par) and self.__input_is_in_case_2(par))
sol = solver.optimize(par)
self.assertAlmostEqual(
sol.profit_ret, 0.01428571
) # would be case 2 solution, because is higher than case 1 solution
self.assertAlmostEqual(sol.profit_man, 0.02857143)
def test_always_case_one(self):
from solver import ModelTwoQuadSolver
par = Parameter(MODEL_2_QUAD,
tau=0.3,
a=0.0005,
s=0.07,
cn=0.67,
cr=0.1,
delta=0.3)
sol = ModelTwoQuadSolver.solve(par, resolution='super high')
print(sol.profit_man, sol.profit_ret)
print(sol)
def test_right_case(self):
solver = ModelTwoNumericalSolver()
#a = np.
par = Parameter(MODEL_2,
tau=.09,
a=0.00146,
s=.04,
cr=.04,
cn=.1,
delta=.7956)
sol = solver.optimize(par)
self.assertAlmostEqual(sol.profit_man, 0.1582925507399)
def case_model_nb_fixed_cn():
a = np.linspace(0.001, 0.02, num=10)
y = []
for act_a in a:
act_a = float(act_a)
par = Parameter(MODEL_2_QUAD, tau=0.09, a=act_a, s=0.4*0.1, cr=0.4*0.1, cn=0.1, delta=.7956)
sol = ModelNBSolver.solve(par, 'low')
print(act_a)
y.append(sol.dec.b)
y.append()
plt.plot(a, y)
plt.show()
def test_something(self):
search = ModelTwoGridSearch()
par = Parameter(MODEL_2_QUAD,
tau=0.09,
a=0.0008163265306122449,
s=0.04000000000000001,
cn=0.1,
cr=0.04000000000000001,
delta=0.7956)
sol = search.search(par)
print(sol.profit_man, sol.dec.wn, sol.dec.pr)
self.assertIsNotNone(sol)
def test_case_1a(self):
solver = ModelOneNumericalSolver()
# this parameter should lead to case one (rho is gte 1)
par = Parameter(MODEL_1, tau=0.1, a=0.005, s=0.0005, cn=0.01)
# self checking the my test input variables..
self.assertTrue(self.__input_is_in_case_1(par))
analyitcal_profit_manufacturer = (
(1 - par.cn)**2 / 8) - (1 / 2) * (1 + par.cn - 2 * par.s) * (
par.tau * par.a)**(1 / 2) + (1 / 2) * par.a * par.tau
sol = solver.optimize(par)
self.assertAlmostEqual(analyitcal_profit_manufacturer, sol.profit_man)
def test_something(self):
from solver import ModelTwoSolver
par = Parameter(MODEL_2_QUAD,
tau=0.09,
a=0.0020408163265306124,
s=0.04000000000000001,
cn=0.1,
cr=0.04000000000000001,
delta=0.7956)
sol = ModelTwoSolver.solve(par)
print(sol)
print(ModelTwoSolver.solve_analytical(par))
def __calc__(self):
p = Factorial
iter = self.__state_iter
proxy = SolverProxy()
for tau, tau_state in iter(p.tau_lh()):
for delta, delta_state in iter(p.delta_lh()):
for cn, cn_state in iter(p.cn_lh()):
for cr, cr_state in iter(p.cr_lh(cn, delta)):
for s, s_state in iter(p.s_lh(cn)):
for a, a_state in iter(p.a_lh()):
# calculate model one
par_n = Parameter(MODEL_1, tau=tau, a=a, s=s, cn=cn)
sol_n = proxy.calculate(par_n)
key_n = self.__key(MODEL_1, tau_state, a_state, s_state, cr_state, cn_state, delta_state)
self._add(key_n, par_n, sol_n)
# calculate model two
par_o = Parameter(MODEL_2, tau=tau, a=a, s=s, cr=cr, cn=cn, delta=delta)
sol_o = proxy.calculate(par_o)
key_o = self.__key(MODEL_2, tau_state, a_state, s_state, cr_state, cn_state, delta_state)
self._add(key_o, par_o, sol_o)
def test_case_2_dec_vars(self):
solver = ModelOneNumericalSolver()
# this args should lead to case two (rho is equal to 1)
par = Parameter(MODEL_1, tau=0.1, a=0.006, s=0.005, cn=0.3)
# self checking the test input variables
self.assertTrue(self.__input_is_in_case_2(par))
sol = solver.optimize(par)
self.assertAlmostEqual(sol.dec.pn, 0.83319444, msg='pn not the same')
self.assertAlmostEqual(sol.dec.wn, (1 / (1 - par.tau)) *
((1 + par.cn) / 2 - (par.tau *
(1 + par.s)) / 2),
msg='wn not the same')
self.assertAlmostEqual(sol.dec.rho, 1.0, msg='rho not the same')
def test_blub(self):
from solver import ModelTwoSolver, ModelNBSolver
import numpy as np
np.seterr(all='ignore')
#par_nb = Parameter(MODEL_NB, tau=0.15, a=0.001, s=0, cn=0.1)
par_nb = Parameter(MODEL_NB,
tau=0.15,
cn=0.1,
s=0.5 * 0.1,
a=0.024329896907216497)
sol = ModelNBSolver.solve(par_nb, resolution='high')
print(sol.profit_man, sol.profit_ret, sol.dec.wn, sol.dec.b, sol.case,
sol.dec.rho)
def test_case_2a(self):
solver = ModelOneNumericalSolver()
# this args should lead to case two (rho is equal to 1)
par = Parameter(MODEL_1, tau=0.1, a=0.006, s=0.005, cn=0.3)
# self checking the test input variables..
# if the following condition is true, it must lead to a case b optimization
self.assertTrue(
self.__input_is_in_case_2(par)
and not self.__input_is_in_case_1(par))
analyitcal_profit_manufacturer = (
(1 - par.cn - par.tau + par.s * par.tau)**2) / (8 * (1 - par.tau))
sol = solver.optimize(par)
self.assertAlmostEqual(analyitcal_profit_manufacturer, sol.profit_man)
def test_instance_a(self):
solver = ModelTwoNumericalSolver()
par = Parameter(MODEL_2,
tau=0.15,
a=0.0013471502590673575,
s=0.05,
cr=0.01,
cn=0.1,
delta=0.85)
dec = solver._optimize_case_one_c(par)
profit_man, profit_ret = solver.calc_profits(par, dec)
sol = Solution(dec, profit_man, profit_ret, '1c')
print(dec.qr - (par.tau / dec.rho) * dec.qn)
print(solver._is_valid(par, sol))
print(profit_man)
