Python error: operands could not be broadcast together with shapes - python

I'm going to do 2D Gaussian fit in Python 3 by:
def gaussian_fit_2_d(data, amplitude, x0, y0, sigma_x, sigma_y, theta, offset):
x0 = np.float64(x0)
y0 = np.float64(y0)
a = (np.cos(theta) ** 2) / (2 * sigma_x ** 2) + (np.sin(theta) ** 2) / (2 * sigma_y ** 2)
b = -np.sin(2 * theta) / (4 * sigma_x ** 2) + np.sin(2 * theta) / (4 * sigma_y ** 2)
c = (np.sin(theta) ** 2) / (2 * sigma_x ** 2) + (np.cos(theta) ** 2) / (2 * sigma_y ** 2)
x = data["x"]
y = data["y"]
f = offset + amplitude * np.exp(-((a * (x - x0)) ** 2 + 2 * b * (x - x0) * (y - y0) + c * (y - y0) ** 2))
return f.ravel()
image = sio.loadmat("File Name")
intensity = np.float64(np. array(image["p1"]))
data_size = intensity.shape
x = np.arange(0, data_size[1])
y = np.arange(0, data_size[0])
# print(data_size)
# print(x)
[X, Y] = np.meshgrid(x, y)
data = {"x": X, "y": Y}
popt, pcov = opt.curve_fit(gaussian_fit_2_d, data, intensity)
,but in the last line I got the following error:
ValueError: operands could not be broadcast together with shapes
(1558,) (41,38)
any idea?

Related

Find the domain of x and y from the intersection line of two surfaces

I have two point cloud and I use scipy.optimize.curve_fit to get the equations representing two surfaces:
z1 = F1(x, y) = (A1 * x ** 2) + (B1 * y ** 2) + (C1 * x * y) + (D1 * x) + (E1 * y) + F1
z2 = F2(x, y) = (A2 * x ** 2) + (B2 * y ** 2) + (C2 * x * y) + (D2 * x) + (E2 * y) + F2
And the equation of their intersection line is:
MY GOAL: Find the domain of x and y
For example: if x0 is in the domain of the intersection line, what is the range of y corresponding to x0?
I have tried some methods with sympy or scipy, but cannot get the domian.

how to pass two arrays into a fuction

I have a function that typically takes in constant args and calculates volatility. I want to pass in a vector of different C's and K's to get an array of volatilities each associated with C[i], K[i]
def vol_calc(S, T, C, K, r, q, sigma):
d1 = (np.log(S / K) + (r - q + 0.5 * sigma ** 2) * T) / (sigma * np.sqrt(T))
vega = (1 / np.sqrt(2 * np.pi)) * np.exp(-q * T) * np.sqrt(T) * np.exp((-si.norm.cdf(d1, 0.0, 1.0) ** 2) * 0.5)
tolerance = 0.000001
x0 = sigma
xnew = x0
xold = x0 - 1
while abs(xnew - xold) > tolerance:
xold = xnew
xnew = (xnew - fx - C) / vega
return abs(xnew)
but if I want to pass two arrays without turning into a nested loop, I thought I could just do:
def myfunction(S, T, r, q, sigma):
for x in K,C:
return same size as K,C
but I can't get it to work
How about this?
def vol_calc(S, T, C, K, r, q, sigma):
import numpy as np
output = np.zeros(len(C))
for num, (c, k) in enumerate(zip(C, K)):
d1 = (np.log(S / k) + (r - q + 0.5 * sigma ** 2) * T) / (sigma * np.sqrt(T))
vega = (1 / np.sqrt(2 * np.pi)) * np.exp(-q * T) * np.sqrt(T) * np.exp((-si.norm.cdf(d1, 0.0, 1.0) ** 2) * 0.5)
tolerance = 0.000001
x0 = sigma
xnew = x0
xold = x0 - 1
while abs(xnew - xold) > tolerance:
xold = xnew
xnew = (xnew - fx - c) / vega
output[num] = abs(xnew)
return output

Gaussian fit: find the X values for a given Y

I did the best fit for my Gaussian curve with Python. Once I have the best fit curve, I would like to know for a given Y value, the correspondent X values.
Since it is a Gaussian curve, I should have two values of X for a given Y ( less than the max value of Y).
How could I do it on Python?
Thank you
Let's take for instance your Gaussian : g(x) = a * exp(-(x - x0) ** 2 / (2 * sigma ** 2)) with a, x0 and sigma being your found parameters from the fit.
And now you want to find y such as y = g(x)
<=> y = a * exp(-(x - x0) ** 2 / (2 * sigma ** 2))
<=> ln|y/a| = -(x - x0) ** 2 / (2 * sigma ** 2))
<=> - ln|y/a| * (2 * sigma ** 2)) =(x - x0) ** 2
<=> sqrt(- ln|y/a| * (2 * sigma ** 2))) = +/- (x - x0)
<=> x = +/- (x0 + sqrt(- ln|y/a| * (2 * sigma ** 2))))
Then, you can use:
def inv_gaus(y, a, x0, sigma):
x = x0 + np.sqrt(-2 * np.log(y / a) * sigma ** 2)
return -x, x
And therefore you have 2 x for one y :)

Integrate Over Multiple Columns in 1 List to Fill Additional List With Same Number Of Columns

I am intending to take a list of random variables and alter a previous list in each column by said random variables. However, for the purpose of my function, each variable must be used in a Gamma function as well as integrated.
x[t] = c * (1 / (2 ** (v / 2) + test[t - 1]) * (gamma((v / 2) + test[t - 1]))) * integrate.\
quad(lambda h: np.exp(-h / 2) * h ** ((v / 2) + test[t - 1] - 1), 0, np.inf)
x[ t ] is an np.zeros((x , y)) list, and test[t - 1] is an np.zeros((x - 1, y)) list
I have filled test[ ] with the appropriate random variables, but I am unable to pass them through this equation to complete the columns of row [ t ] in x
When I try to run my current code, I receive:
File "C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\scipy\integrate\quadpack.py", line 450, in _quad
return _quadpack._qagie(func,bound,infbounds,args,full_output,epsabs,epsrel,limit)
TypeError: only size-1 arrays can be converted to Python scalars
Is there a different special function which allows me to use each column's variable to solve for my desired x[ t ]?
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import stats
import mpmath as mp
import scipy.integrate as integrate
from scipy.special import gamma
T = 1
beta = 0.5
x0 = 0.05
q = 0
mu = x0 - q
alpha = - (2 - beta) * mu
sigma0 = 0.1
sigma = (2 - beta) * sigma0
b = - ((1 - beta) / (2 * mu) * (sigma0 ** 2))
simulations = 100
M = 50
dt = T / M
def srd_sampled_nxc2():
x = np.zeros((M + 1, simulations))
x[0] = x0
test = np.zeros((M, simulations))
for t in range(1, M + 1):
v = 4 * b * alpha / sigma ** 2
c = (sigma ** 2 * (1 - np.exp(-alpha * dt))) / (4 * alpha)
nc = np.exp(-alpha * dt) / c * x[t - 1]
if v > 1:
x[t] = c * ((np.random.standard_normal(simulations) + nc ** 0.5) ** 2 + mp.nsum(
lambda i: np.random.standard_normal(simulations) ** 2, [0, v - 1]))
else:
max_array = []
nc_over_2 = [l / 2 for l in nc]
for p in range(simulations):
sump = []
poisson_start = 0
while poisson_start <= 1:
x_i = sum(-np.log(np.random.uniform(0, 1, simulations)) / nc_over_2)
sump.append(
x_i
)
poisson_start += x_i
x_n = max(sump)
max_array.append(
x_n
)
sump = []
test[t - 1] = max_array
x[t] = c * (1 / (2 ** ((v / 2) + test[t - 1])) * (gamma((v / 2) + test[t - 1]))) * integrate.\
quad(lambda h: np.exp(-h / 2) * h ** ((v / 2) + test[t - 1] - 1), 0, np.inf)
max_array = []
return x
Ultimately ended up finding a workaround which is simple to implement:
else:
max_array = []
for p in range(simulations):
k = nc[t - 1, p]
lam = k / 2
poisson_samp = 0
while poisson_samp <= 1:
x_i = -math.log(np.random.uniform(0, 1)) / lam
max_array.append(
x_i
)
poisson_samp += x_i
test[t - 1, p] = len(max_array) - 1
max_array.clear()
for f in range(simulations):
n = test[t - 1, f]
z = integrate.quad(lambda h: np.exp(-h / 2) * h ** ((v / 2) + n - 1), 0, 1)
new[t - 1, f] = z[0]
x[t] = c * (1 / (2 ** ((v / 2) + test[t - 1]) * (gamma((v / 2) + test[t - 1]))) * new[0])
The only real problem is the shrinkage of x[t] which leads to dividing by zero--just a formula problem.

ode integration in python versus mathematica results

Edit:
So I found out that NDSolve for ODE is using Runge Kutta to solve the equations.
How can I use the Runge Kutta method on my python code to solve the ODE I have below?
From my post on text files with float entries, I was able to determine that python and mathematica immediately start diverging with a tolerance of 10 to the negative 6.
End Edit
For last few hours, I have been trying to figure out why my solutions in Mathematica and Python differ by 5000 something km.
I am led to believe one program has a higher error tolerance when simulating over millions of seconds in flight time.
My question is which program is more accurate, and if it isn't python, how can I adjust the precision?
With Mathematica, I am less than 10km away from L4 where as with Python I am 5947 km away.
The codes are listed below:
Python
import numpy as np
from scipy.integrate import odeint
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from numpy import linspace
from scipy.optimize import brentq
me = 5.974 * 10 ** (24) # mass of the earth
mm = 7.348 * 10 ** (22) # mass of the moon
G = 6.67259 * 10 ** (-20) # gravitational parameter
re = 6378.0 # radius of the earth in km
rm = 1737.0 # radius of the moon in km
r12 = 384400.0 # distance between the CoM of the earth and moon
d = 300 # distance the spacecraft is above the Earth
pi1 = me / (me + mm)
pi2 = mm / (me + mm)
mue = 398600.0 # gravitational parameter of earth km^3/sec^2
mum = G * mm # grav param of the moon
mu = mue + mum
omega = np.sqrt(mu / (r12 ** 3))
nu = -np.pi / 4 # true anomaly pick yourself
xl4 = r12 / 2 - 4671 # x location of L4
yl4 = np.sqrt(3) / 2 * r12 # y
print("The location of L4 is", xl4, yl4)
# Solve for Jacobi's constant
def f(C):
return (omega ** 2 * (xl4 ** 2 + yl4 ** 2) + 2 * mue / r12 + 2 * mum / r12
+ 2 * C)
c = brentq(f, -5, 0)
print("Jacobi's constant is",c)
x0 = (re + 200) * np.cos(nu) - pi2 * r12 # x location of the satellite
y0 = (re + 200) * np.sin(nu) # y location
print("The satellite's initial position is", x0, y0)
vbo = (np.sqrt(omega ** 2 * (x0 ** 2 + y0 ** 2) + 2 * mue /
np.sqrt((x0 + pi2 * r12) ** 2 + y0 ** 2) + 2 * mum /
np.sqrt((x0 - pi1 * r12) ** 2 + y0 ** 2) + 2 * -1.21))
print("Burnout velocity is", vbo)
gamma = 0.4678 * np.pi / 180 # flight path angle pick yourself
vx = vbo * (np.sin(gamma) * np.cos(nu) - np.cos(gamma) * np.sin(nu))
# velocity of the bo in the x direction
vy = vbo * (np.sin(gamma) * np.sin(nu) + np.cos(gamma) * np.cos(nu))
# velocity of the bo in the y direction
print("The satellite's initial velocity is", vx, vy)
# r0 = [x, y, 0]
# v0 = [vx, vy, 0]
u0 = [x0, y0, 0, vx, vy, 0]
def deriv(u, dt):
return [u[3], # dotu[0] = u[3]
u[4], # dotu[1] = u[4]
u[5], # dotu[2] = u[5]
(2 * omega * u[4] + omega ** 2 * u[0] - mue * (u[0] + pi2 * r12) /
np.sqrt(((u[0] + pi2 * r12) ** 2 + u[1] ** 2) ** 3) - mum *
(u[0] - pi1 * r12) /
np.sqrt(((u[0] - pi1 * r12) ** 2 + u[1] ** 2) ** 3)),
# dotu[3] = that
(-2 * omega * u[3] + omega ** 2 * u[1] - mue * u[1] /
np.sqrt(((u[0] + pi2 * r12) ** 2 + u[1] ** 2) ** 3) - mum * u[1] /
np.sqrt(((u[0] - pi1 * r12) ** 2 + u[1] ** 2) ** 3)),
# dotu[4] = that
0] # dotu[5] = 0
dt = np.linspace(0.0, 6.0 * 86400.0, 2000000.0) # secs to run the simulation
u = odeint(deriv, u0, dt)
x, y, z, x2, y2, z2 = u.T
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.plot(x, y, z, color = 'r')
# adding the Lagrange point
phi = np.linspace(0, 2 * np.pi, 100)
theta = np.linspace(0, np.pi, 100)
xm = 2000 * np.outer(np.cos(phi), np.sin(theta)) + xl4
ym = 2000 * np.outer(np.sin(phi), np.sin(theta)) + yl4
zm = 2000 * np.outer(np.ones(np.size(phi)), np.cos(theta))
ax.plot_surface(xm, ym, zm, color = '#696969', linewidth = 0)
ax.auto_scale_xyz([-8000, 385000], [-8000, 385000], [-8000, 385000])
# adding the earth
phi = np.linspace(0, 2 * np.pi, 100)
theta = np.linspace(0, np.pi, 100)
xm = 2000 * np.outer(np.cos(phi), np.sin(theta))
ym = 2000 * np.outer(np.sin(phi), np.sin(theta))
zm = 2000 * np.outer(np.ones(np.size(phi)), np.cos(theta))
ax.plot_surface(xm, ym, zm, color = '#696969', linewidth = 0)
ax.auto_scale_xyz([-8000, 385000], [-8000, 385000], [-8000, 385000])
plt.show()
# The code below finds the distance between path and l4
my_x, my_y, my_z = (xl4, yl4, 0.0)
delta_x = x - my_x
delta_y = y - my_y
delta_z = z - my_z
distance = np.array([np.sqrt(delta_x ** 2 + delta_y ** 2 + delta_z ** 2)])
minimum = np.amin(distance)
print("Closet approach to L4 is", minimum)
Mathematica
ClearAll["Global`*"];
me = 5.974*10^(24);
mm = 7.348*10^(22);
G = 6.67259*10^(-20);
re = 6378;
rm = 1737;
r12 = 384400;
\[Pi]1 = me/(me + mm);
\[Pi]2 = mm/(me + mm);
M = me + mm;
\[Mu]1 = 398600;
\[Mu]2 = G*mm;
\[Mu] = \[Mu]1 + \[Mu]2;
\[CapitalOmega] = Sqrt[\[Mu]/r12^3];
\[Nu] = -\[Pi]/4;
xl4 = 384400/2 - 4671;
yl4 = Sqrt[3]/2*384400 // N;
Solve[\[CapitalOmega]^2*(xl4^2 + yl4^2) + 2 \[Mu]1/r12 +
2 \[Mu]2/r12 + 2*C == 0, C]
x = (re + 200)*Cos[\[Nu]] - \[Pi]2*r12 // N
y = (re + 200)*Sin[\[Nu]] // N
{{C -> -1.56824}}
-19.3098
-4651.35
vbo = Sqrt[\[CapitalOmega]^2*((x)^2 + (y)^2) +
2*\[Mu]1/Sqrt[(x + \[Pi]2*r12)^2 + (y)^2] +
2*\[Mu]2/Sqrt[(x - \[Pi]1*r12)^2 + (y)^2] + 2*(-1.21)]
10.8994
\[Gamma] = 0.4678*Pi/180;
vx = vbo*(Sin[\[Gamma]]*Cos[\[Nu]] - Cos[\[Gamma]]*Sin[\[Nu]]);
vy = vbo*(Sin[\[Gamma]]*Sin[\[Nu]] + Cos[\[Gamma]]*Cos[\[Nu]]);
r0 = {x, y, 0};
v0 = {vx, vy, 0}
{7.76974, 7.64389, 0}
s = NDSolve[{x1''[t] -
2*\[CapitalOmega]*x2'[t] - \[CapitalOmega]^2*
x1[t] == -\[Mu]1/((Sqrt[(x1[t] + \[Pi]2*r12)^2 +
x2[t]^2])^3)*(x1[t] + \[Pi]2*
r12) - \[Mu]2/((Sqrt[(x1[t] - \[Pi]1*r12)^2 +
x2[t]^2])^3)*(x1[t] - \[Pi]1*r12),
x2''[t] +
2*\[CapitalOmega]*x1'[t] - \[CapitalOmega]^2*
x2[t] == -\[Mu]1/(Sqrt[(x1[t] + \[Pi]2*r12)^2 + x2[t]^2])^3*
x2[t] - \[Mu]2/(Sqrt[(x1[t] - \[Pi]1*r12)^2 + x2[t]^2])^3*
x2[t], x3''[t] == 0, x1[0] == r0[[1]], x1'[0] == v0[[1]],
x2[0] == r0[[2]], x2'[0] == v0[[2]], x3[0] == r0[[3]],
x3'[0] == v0[[3]]}, {x1, x2, x3}, {t, 0, 1000000}];
ParametricPlot3D[
Evaluate[{x1[t], x2[t], x3[t]} /. s], {t, 0, 10*24*3600},
PlotStyle -> {Red, Thick}]
g1 = ParametricPlot3D[
Evaluate[{x1[t], x2[t], x3[t]} /. s], {t, 0, 5.75*3600*24},
PlotStyle -> {Red},
PlotRange -> {{-10000, 400000}, {-10000, 400000}}];
g2 = Graphics3D[{Blue, Opacity[0.6], Sphere[{-4671, 0, 0}, re]}];
g3 = Graphics3D[{Green, Opacity[0.6], Sphere[{379729, 0, 0}, rm]}];
g4 = Graphics3D[{Black, Sphere[{xl4, yl4, 0}, 2000]}];
Show[g2, g1, g3, g4, Boxed -> False]
(*XYdata=Flatten[Table[Evaluate[{x1[t],x2[t],x3[t]}/.s],{t,5.5*24*\
3600,5.78*24*3600,1}],1];
X1Y1data=Flatten[Table[Evaluate[{x1'[t],x2'[t],x3'[t]}/.s],{t,5.5*24*\
3600,5.78*24*3600,1}],1];
SetDirectory[NotebookDirectory[]];
Export["OrbitData.txt",XYdata,"CSV"];
Export["OrbVeloc.txt",X1Y1data,"CSV"];*)
If at this point your problem has reduced to just wanting to use Runge-Kutta, you can for example replace this line:
u = odeint(deriv, u0, dt)
with something like this:
#reverse the order of arguments
def deriv2(t,u):
return deriv(u,t)
# initialize a 4th order Runge-Kutta ODE solver
solver = ode(deriv2).set_integrator('dopri5')
solver.set_initial_value(u0)
u = np.empty((len(dt), 6))
u[0,:] = u0
for ii in range(1,len(dt)):
u[ii] = solver.integrate(dt[ii])
(+obviously replace the odeint import with ode).
Note that this is significantly slower for this type of ODE.
To use the dop853, use solver.set_integrator('dop853').
I re-wrote the def deriv part of the ode and it works now! So the Mathematica plot and the Python agree.
def deriv(u, dt):
return [u[3], # dotu[0] = u[3]
u[4], # dotu[1] = u[4]
u[5], # dotu[2] = u[5]
(2 * omega * u[4] + omega ** 2 * u[0] - mue * (u[0] + pi2 * r12) /
np.sqrt(((u[0] + pi2 * r12) ** 2 + u[1] ** 2) ** 3) - mum *
(u[0] - pi1 * r12) /
np.sqrt(((u[0] - pi1 * r12) ** 2 + u[1] ** 2) ** 3)),
# dotu[3] = that
(-2 * omega * u[3] + omega ** 2 * u[1] - mue * u[1] /
np.sqrt(((u[0] + pi2 * r12) ** 2 + u[1] ** 2) ** 3) - mum * u[1] /
np.sqrt(((u[0] - pi1 * r12) ** 2 + u[1] ** 2) ** 3)),
# dotu[4] = that
0] # dotu[5] = 0
One way to check the accuracy could be this:
take a time where the 2 trajectories are sufficiently different from one another (the last time/point in the trajectory for example). Use this as the new starting point and integrate back in time to the initial time (or, if you prefer, integrate forward in time, but reverse the velocities of all bodies present in the simulation). The final point in this simulation should be (close to) your initial point in the original simulation. By comparing how close you actually get to your original initial point, you can judge how good python vs mathematica solver routines are. My guess is that the Mathematica routines are much better, so all this Python thing is a waste of time.
Another way could be this:
With Mathematica what you get is an interpolating function which you can symbolically derive and then numerically evaluate its derivatives. Plug these values in the ODE at different points and see if they satisfy the ODE. Do something similar in python and compare the results.

Categories

Resources