arrays cannot be broadcast to a common size

[stochasticTarget]

I suddenly had a problem using ode without modifying the differential operator.
I use DataFrames, ODE, DataArrays. Since yesterday, without any update or checkout I suddenly obtain from

t = linspace(0,300,30001)# time
result=ode23((t,x)->D3(t,x,parameters),t,inits);

where

• parameters = 15-element Array{Float64,1}
• inits = 3-element Array{Float64,1}
• D3 = (generic function with 1 method) takes (t,inits,parameters) and return an array same size as inits.
  the following result:

arrays cannot be broadcast to a common size
at In[17]:3
 in broadcast_shape at broadcast.jl:37
 in ##broadcast_T_/#233 at broadcast.jl:187
 in ./ at broadcast.jl:288
 in ode23 at /home/adrien/.julia/v0.2/ODE/src/ODE.jl:68

Any conflicting issue with DataArrays or DataFrames?

[tomasaschan]

On line 67-68 I find this:

s1 = F(t,y) 
r = norm(s1./max(abs(y), threshold), Inf) + realmin() # TODO: fix type bug in max()

so it could be the type bug mentioned there. Would you mind sharing the exact definition of D3, as well as your setup code? It's pretty difficult to try to reproduce the error without it.

[stochasticTarget]

The code might not be efficient : it is designed also for teaching.
Here is D3:

function D3(t,X,p)   
    #state space variables
    om = X[1]
    la = X[2]
    de = X[3]
    #parameters
    nu = p[1]
    alpha = p[2]
    beta = p[3]
    gamma = p[4]
    delta = p[5]
    r = p[6]
    c_1 = p[7]
    c_2 = p[8]
    eta_p = p[9]
    m = p[10]
    par_phi = p[11:12]
    par_kappa = p[13:15]  
    # Define the Philips curve function and its parameters
    fun_Phi(x, p_Phi)=-p_Phi[1] + p_Phi[2]/(1-x)^2
    # Define the investment function and its parameters
    fun_Kappa(x,p_Kap)=p_Kap[1] + p_Kap[2]*exp(p_Kap[3]*x) 
    #additional values
    pi = 1 - om - r*de
    y_d = c_1 * om + c_2 * de + fun_Kappa(pi, par_kappa)
    inflation_rate = gamma * (m * om - 1)
    #dynamics
    d_om = om * (fun_Phi(la, par_phi) - alpha - (1-eta_p) * inflation_rate)
    d_la = la * (fun_Kappa(pi, par_kappa)/nu - alpha - beta - delta)
    d_de = de * (r + delta -inflation_rate - fun_Kappa(pi, par_kappa)/nu) + fun_Kappa(pi, par_kappa) - (1-om)*y_d 

    return([d_om;d_la;d_de])
end

And settling parameters :

# Initialise model
# Define the direct parameters of the dynamical system
nu = 3                              # capital-to-output ratio
alpha = 0.025                  # productivity growth rate
beta = 0.02                      # population growth rate
gamma = 0.05                 # markup adjustement speed and worst deflation rate
delta = 0.01                     # depreciation rate of capital
r = 0.03                            # interest rate on deposit/debt
c_1 = 1 #0.97                  # propensity to consume out of wage
c_2 = 0 #0.015#0.02       # propensity to consume out of wealth
eta_p = 0.85                    # rigidity factor of inflation in wage progression
m = 1.1931                      # target markup factor, m-1 = target profit rate
par_phi_0 = 0.04/(1-0.04^2)
par_phi_1 = (0.04^3)/(1-0.04^2)
par_kappa_0 = -0.0065
par_kappa_1 = exp(-5)
par_kappa_2 = 20
####TIME AND INITIAL STATE
Omega_0 = 0.75                 # initial expected wage share
Lambda_0 = 0.75                # employment rate
Debt_0 = 0.2                       # expected debt share
####ARRAYS
inits=[Omega_0,Lambda_0,Debt_0];
parameters = [nu; alpha; beta; gamma; delta; r; c_1; c_2; eta_p; m;
    par_phi_0; par_phi_1; par_kappa_0; par_kappa_1; par_kappa_2]

[acroy]

Which version of ODE.jl are you using? Try to interchange t and inits in your call to ode23, i.e.,

result=ode23((t,x)->D3(t,x,parameters),inits,t);

[stochasticTarget]

Pkg.status() provides 0.1.0+ master
I followed your change. That compiles with the change thanks! Is it a recent change in ODE.jl ?

Is that the same for other ODE functions : ode4 , ode4s, etc?
I thus recommend Pkg.checkout("ODE") and continue coding with the parameters order you suggest.

[acroy]

The change happened with #14, which was merged in March. Did you switch to master recently?
All functions in ODE.jl use the same API, so it should also work with the other solvers.