def calc_S_mat():
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
results = read_results(constants)
sol_k, krashade = calc_sol_k()
suc2, mig1, mig1_phos, X = sol_k
t_span, t_eval, y0 = ode_info
best_coefficients, min_cost_funk, min_viktad_cost_funk = get_min_cost(results)
Kinetic_constants = best_coefficients
s_suc2 = np.zeros((num_tidsteg, num_coefficient))
s_mig1 = np.zeros((num_tidsteg, num_coefficient))
s_mig1_phos = np.zeros((num_tidsteg, num_coefficient))
s_X = np.zeros((num_tidsteg, num_coefficient))
dy = np.zeros([num_tidserier, num_tidsteg, 1])
for i in range(num_coefficient):
d_Kinetic_constants = Kinetic_constants.copy()
d_Kinetic_constants[i] = d_Kinetic_constants[i] + h
d_solv = integrate.solve_ivp(fun=lambda t, y: model(vald_modell, t, y, d_Kinetic_constants), t_span=t_span, y0=y0,
method="RK45", t_eval=t_eval)
d_solv_k = np.zeros([num_tidserier, num_tidsteg, 1])
d_solv_k[:, :, 0] = d_solv.y
dy_new = dy.copy()
dy_new[:] = d_solv_k
s_suc2[:, i] = np.transpose((dy_new[0] - suc2) / h)
s_mig1[:, i] = np.transpose((dy_new[1] - mig1) / h)
s_mig1_phos[:, i] = np.transpose((dy_new[2] - mig1_phos) / h)
s_X[:, i] = np.transpose((dy_new[3] - X) / h)
S = np.array([s_suc2, s_mig1, s_mig1_phos, s_X])
return S
def main_plot_optimering():
time_points, data_concentration = data()
constants, ode_info, data_info = model_info(time_points)
results = read_results(constants)
coefficients, min_cost_funk, min_viktad_cost_funk = get_min_cost(results)
sol_k = solve_ODE(coefficients, constants, ode_info)
mat_r = calc_residual(sol_k, constants, data_concentration, data_info)
plot_all(constants, sol_k, mat_r, data_concentration, time_points)
def fix_invertibility(matrix):
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
indent_mat = np.identity(num_coefficient)
eigenvalues = np.linalg.eigvals(matrix)
min_nonpos_eigenvalue = np.amin(eigenvalues)
gamma = -(min_nonpos_eigenvalue - h)
fixed_matrix = matrix + gamma * indent_mat
return fixed_matrix
def main():
time_points, data_concentration = data()
constants, ode_info, data_info = model_info(time_points)
for i in range(100):
print("runda: " + str(i))
k_array = guess_k_array(i)
start_point(k_array, constants, data_concentration, data_info,
ode_info)
results = iteration(k_array, constants, data_concentration, data_info,
ode_info)
save_results(results, constants)
def corr():
H_inv = h_inverse()
Covariance = H_inv
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
v0 = np.zeros((num_coefficient, 1))
Correlation = np.zeros((num_tidserier, num_coefficient, num_coefficient))
for i in range(num_tidserier):
v = v0.copy()
v = np.sqrt(np.diag(Covariance[i, :, :]))
outer_v = np.outer(v, v)
Correlation[i, :, :] = Covariance[i, :, :] / outer_v
return Correlation
def calc_sol_k():
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
results = read_results(constants)
best_coefficients, min_cost_funk, min_viktad_cost_funk = get_min_cost(results)
t_span, t_eval, y0 = ode_info
sol = integrate.solve_ivp(fun=lambda t, y: model(vald_modell, t, y, best_coefficients), t_span=t_span, y0=y0, method="RK45",
t_eval=t_eval)
sol_k = np.empty([num_tidserier, num_tidsteg, 1])
krashade = False
try:
sol_k[:, :, 0] = sol.y
except ValueError:
krashade = True
return sol_k, krashade
def RMS():
S = calc_S_mat()
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
results = read_results(constants)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
best_coefficients, min_cost_funk, min_viktad_cost_funk = get_min_cost(results)
Kinetic_constants = best_coefficients
sol_k, krashade = calc_sol_k()
suc2, mig1, mig1_phos, X = sol_k
Model_values = np.transpose(np.array([suc2, mig1, mig1_phos, X]))
rms = np.zeros((2, num_coefficient))
S_square = np.power(S, 2)
model_square = np.power(Model_values, 2).reshape(num_tidsteg, num_tidserier)
for j in range(2):
for i in range(num_coefficient):
K_square = np.power(Kinetic_constants, 2)
rms[j, i] = math.sqrt((1/num_tidsteg)*np.sum((S_square[j, :, i]*K_square[i]/model_square[:, j]), axis=0))
return rms
def var_K():
H_inv = h_inverse()
time_points, data_conc = data()
constants, ode_info, data_info = model_info(time_points)
vald_modell, num_coefficient, num_tidserier, num_tidsteg, h = constants
results = read_results(constants)
best_coefficients, min_cost_funk, min_viktad_cost_funk = get_min_cost(
results)
Kinetic_constants = best_coefficients
Var_K = np.zeros((num_tidserier, num_coefficient, 1))
for i in range(num_tidserier):
c = np.sqrt(np.diag(H_inv[i, :, :]).reshape(num_coefficient, 1))
for l in np.diag(H_inv[i]):
if l < 0:
print('COV ej definierad')
break
for j in range(num_coefficient):
if Kinetic_constants[j] == 0:
Var_K[i, j] = 0
else:
Var_K[i, j] = c[j] / Kinetic_constants[j]
return Var_K