e2initialmodel_sgd

usage: 
	This program makes initial models using a (kind of) stochastic gradient descent approach. It is recommended that the box size of particles is around 100. 
	[prog] --ptcls <particle stack>

Option	Type	Description
--version	None	show program's version number and exit
--path	str	Path to write initial model output. Default is initmodel_XX
--ptcls	str	Class average or particles input.
--sym	str	Specify symmetry - choices are: c<n>, d<n>, h<n>, tet, oct, icos
--batchsize	int	Batch size of stochastic gradient desent. N particles are randomly selected to generate an initial model at each step.
--niter	int	Number of iterations
--ntry	int	The number of different initial models to generate in search of a good one
--learnrate	float	Learning rate. i.e. how much the initial model changes toward the gradient direction in each iteration. Range from 0.0~1.0. Default is 0.3
--lrdecay	float	Learning rate multiplier after each iteration.
--addnoise	float	Add noise on particles at each iteration. Stablize convergence for some reason.
--shrink	int	shrinking factor
--setsf	str
--ref	str
--writetmp	None	Write output for each iteration
--fullcov	None	Assume the input particles covers most of the orientation of the model. This gives better performance when the model is relatively feature-less, but is more likely to fail when there are incorrect particles in the input.
--threads	int	threads
--verbose, -v	int	Verbose
--targetres	float	Target resolution