Hi, You'll probably see that this is also the difference between the scipy.optimize.leastsq() Levenberg-Marquardt optimisation function and the other scipy.optimize functions. There scipy.optimise interfaces are not standardised. However if you look at minfx Levenberg-Marquardt interface:
http://home.gna.org/minfx/minfx.levenberg_marquardt-module.html You will see the dfunc argument which should be the Jacobian returning function. The Levenberg-Marquardt optimisation is quite unique in requiring additional information about the Jacobian. None of the other algorithms in minfx or scipy require this. Regards, Edward On 26 August 2014 13:12, <[email protected]> wrote: > Author: tlinnet > Date: Tue Aug 26 13:12:14 2014 > New Revision: 25285 > > URL: http://svn.gna.org/viewcvs/relax?rev=25285&view=rev > Log: > Got the method of 'Steepest descent' to work properly, by specifying the > Jacobian correctly. > > The Jacobian was derived according to the chi2 function. > > The key point was to evaluate the two derivative arrays for all times points, > and > and then sum each of the two arrays together, before constructing the > Jacobian. > > This clearly shows the difference between minfx and scipy.optimize.leastsq. > > scipy.optimize.leastsq takes as input a function F(x0), which should return > the array > of weighted differences between function value and measured values: > " 1. / self.errors * (self.calc_exp(self.times, *params) - self.values) " > > This will be an array with number of elements 'i' corresponding to number of > elements. > scipy.optimize.leastsq then internally evaluates the sum of squares -> sum[ > (O - E)**2 ], > and minimises this. This is the chi2. > > Minfx requires the function to minimise before hand. > So, the "func" should be chi2. > > Then the dfunc, and d2func, should be derivative of chi2, bum all elements in > the array should > still be summed together. > > task #7822(https://gna.org/task/index.php?7822): Implement user function to > estimate R2eff and associated errors for exponential curve fitting. > > Modified: > trunk/specific_analyses/relax_disp/estimate_r2eff.py > > Modified: trunk/specific_analyses/relax_disp/estimate_r2eff.py > URL: > http://svn.gna.org/viewcvs/relax/trunk/specific_analyses/relax_disp/estimate_r2eff.py?rev=25285&r1=25284&r2=25285&view=diff > ============================================================================== > --- trunk/specific_analyses/relax_disp/estimate_r2eff.py (original) > +++ trunk/specific_analyses/relax_disp/estimate_r2eff.py Tue Aug 26 > 13:12:14 2014 > @@ -134,18 +134,6 @@ > # Set algorithm. > self.min_algor = min_algor > > - # Newton does not work. > - #self.min_algor = 'newton' > - > - # Newton-CG does not work. > - #self.min_algor = 'Newton-CG' > - > - # Also not work. > - #self.min_algor = 'Steepest descent' > - > - # Also not work.# > - #self.min_algor = 'Fletcher-Reeves' > - > # Define if constraints should be used. > self.constraints = constraints > > @@ -164,7 +152,6 @@ > self.b = array([ 0., -200., 0.]) > > else: > - self.min_algor = 'simplex' > self.min_options = () > self.A = None > self.b = None > @@ -305,16 +292,15 @@ > > # Make partial derivative, with respect to r2eff. > # d_chi2_d_r2eff = 2.0*i0*times*(-i0*exp(-r2eff*times) + > values)*exp(-r2eff*times)/errors**2 > - d_chi2_d_r2eff = 2.0 * i0 * self.times * ( -i0 * exp( -r2eff * > self.times) + self.values) * exp( -r2eff * self.times ) / self.errors**2 > + d_chi2_d_r2eff = sum( 2.0 * i0 * self.times * ( -i0 * exp( -r2eff * > self.times) + self.values) * exp( -r2eff * self.times ) / self.errors**2 ) > > # Make partial derivative, with respect to i0. > # d_chi2_d_i0 = -2.0*(-i0*exp(-r2eff*times) + > values)*exp(-r2eff*times)/errors**2 > - d_chi2_d_i0 = - 2.0 * ( -i0 * exp( -r2eff * self.times) + > self.values) * exp( -r2eff * self.times) / self.errors**2 > + d_chi2_d_i0 = sum ( - 2.0 * ( -i0 * exp( -r2eff * self.times) + > self.values) * exp( -r2eff * self.times) / self.errors**2 ) > > # Define Jacobian as m rows with function derivatives and n columns > of parameters. > - jacobian_matrix = transpose(array( [d_chi2_d_r2eff , d_chi2_d_i0] ) ) > - > - #print jacobian_matrix > + #jacobian_matrix = transpose(array( [d_chi2_d_r2eff , d_chi2_d_i0] ) > ) > + jacobian_matrix = array( [d_chi2_d_r2eff , d_chi2_d_i0] ) > > # Return Jacobian matrix. > return jacobian_matrix > @@ -439,7 +425,7 @@ > # 'minfx' > # 'scipy.optimize.leastsq' > # 'scipy.optimize.fmin_cg' > -def estimate_r2eff(spin_id=None, ftol=1e-15, xtol=1e-15, maxfev=10000000, > factor=100.0, method='scipy.optimize.leastsq', verbosity=1): > +def estimate_r2eff(spin_id=None, ftol=1e-15, xtol=1e-15, maxfev=10000000, > factor=100.0, method='minfx', verbosity=1): > """Estimate r2eff and errors by exponential curve fitting with > scipy.optimize.leastsq. > > scipy.optimize.leastsq is a wrapper around MINPACK's lmdif and lmder > algorithms. > @@ -782,7 +768,24 @@ > x0 = asarray( E.estimate_x0_exp() ) > > # Set the min_algor. > - E.set_settings_minfx(min_algor='simplex') > + #min_algor='simplex' > + > + # Steepest descent uses the gradient. > + min_algor = 'Steepest descent' > + max_iterations = 1000 > + > + # Newton does not work. > + # min_algor = 'newton' > + > + # Newton-CG does not work. > + # min_algor = 'Newton-CG' > + > + > + > + # Also not work.# > + # min_algor = 'Fletcher-Reeves' > + > + E.set_settings_minfx(min_algor=min_algor, max_iterations=max_iterations) > > # Define function to minimise for minfx. > if match('^[Ss]implex$', E.min_algor): > @@ -791,7 +794,7 @@ > dfunc = None > d2func = None > else: > - func = E.func_exp_chi2 > + func = E.func_exp > dfunc = E.func_exp_grad > d2func = E.func_exp_hess > > > > _______________________________________________ > relax (http://www.nmr-relax.com) > > This is the relax-commits mailing list > [email protected] > > To unsubscribe from this list, get a password > reminder, or change your subscription options, > visit the list information page at > https://mail.gna.org/listinfo/relax-commits _______________________________________________ relax (http://www.nmr-relax.com) This is the relax-devel mailing list [email protected] To unsubscribe from this list, get a password reminder, or change your subscription options, visit the list information page at https://mail.gna.org/listinfo/relax-devel
