2003-08-18  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlpwrite.m: Use use_ard in hyperest.

2003-05-15  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* Release 1.0: Major cleanup, some bug fixes. All functions
	  checked, made more independent, and functions not used were
	  removed. Ready for public release.

2002-04-18  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlpwrite.m: Added checking if class(net)=='network'

2001-05-24  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlpwrite.m: Support fbmtoolbox style networks.

2001-01-04  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlpread.m: Fixed two typos Sigma->Sigmas

2000-07-12  Simo Srkk  <Simo.Sarkka@hut.fi>

	* fbmthin.m: Cell arrays of size 1 are now also thinned.

2000-05-25  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmgppred.m: Added linear term to covariance calculation.

	* fbmgpread.m: Read linear sigmas too.

2000-04-28  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlppred.m: Added fast method (3 times faster with 19x20x1 MLP).

2000-04-18  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmgpread.m: Added type field.

	* fbmmlpread.m: Ditto.

2000-04-17  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlpwrite.m: outputSigmas->noiseSigmas, outputHyper->noiseHyper

	* fbmgppred.m: Use fbmthin.
		       + outputSigmas->noiseSigmas, outputHyper->noiseHyper

	* fbmmlpdiag.m: Ditto.

	* fbmmlppred.m: Ditto.

	* fbmmlpsim.m: Ditto.

	* fbmthin.m: One general thinning function.

	* fbmgpthin.m: Removed.

	* fbmmlpthin.m: Ditto.

	* fbmjoin.m: New function for joining fbm models.

	* fbmgpthin.m: nlast fix + accept arrays of fbm structures

2000-04-13  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmmlpread.m: If using nlast don't warn about not reading entire
	                file.

	* fbmmlpwrite.m: Handle classification models too.

2000-04-05  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* fbmgppred.m: First working version.

	* fbmmlpthin.m: Check size only once.

	* fbmgpthin.m: Add nlast. Check size only once. Fixes.

	* fbmmlppred.m: Avoid using structures in loops because structures
	                are slow. Extra variables do not use any extra
	                memory because they are pointers to original data.
	                + some cosmetic surgery.
			+ use nlast.

2000-04-04  Aki Vehtari  <Aki.Vehtari@hut.fi>

	* private/split_s.m: Added check for variances as classification
	                     (logit/softmax) models do not have variances.

	* private/join_s.m: Ditto.

2000-04-04  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlpsim.m: Fixed bug that prohibited quantile
		       calculation for multiple outputs.
	
1999-11-09  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlpread.m: The function reads now only the needed
	                records and skips the rest.

1999-11-03  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlppred.m: Added arguments nrnd and nlast. Changed
	                function to return sample from the true
	                predictive distribution instead of distribution
	                of output means.

1999-10-28  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlp2com.m: Added argument nlast

1999-10-27  Simo Srkk  <ssarkka@zeus.lce.hut.fi>

	* fbmmlp2com.m: Function changed to use new function
	                fbmmlp2nnet(). Also added support for creating
	                one Committee from multiple MCMC-simulations.
	

