1.6-3 (24 Aug 2011)
---
Move back to L-based convergence with theta check
Predict takes simple topic matrices and prints dispersion
Remove some unused switches (sqp, nef, etc)
Cleaner summary print-out and return object

1.6-2 (10 Aug 2011)
---
Fix bug on non-reporting of dispersion
New defaults on initialization
Add hidden option for init tmax
Allow for removal of empty docs
Change to matrix return from summary
Check for zeros in initheta

1.6-1 (1 Aug 2011)
---
Small update to we8there and congress data examples
Added example of lm in fwd regression for predict
Add lift in return object from summary for topics

1.6 (8 June 2011)
---
Added the option for basic mixture modeling with admix=FALSE
New tpxSelect function, replacing tpxBF.  Cleaner initialization.
Fixed a small bug in BF calculation; now appears very solid.
Fixed stemming in we8there
Move to `lift' terminology for summary
Add dispersion estimation and summary
Move to concentration=1+shape parameterization
	and introduce capability for full matrix shape

1.5-4 (16 April 2011)
---
Fixed a strange bug with summary for factor response input.  Letting slam
	specify its own `counts' matrix dimensions on imput seems to solve things.
Fixed an ommited plot (polychotomous classification) and added example for mnlm
Fixed a bug in coefficient initialization.
Minor fixes in logistic regression summary and plot.
Add a coefficient bound to avoid infinite denominator under almost complete separation.

1.5 (8 April 2011)
---
Switched to independent lambda penalties for MAP estimation.
mnlm `lambda' input argument is now called `penalty'.
A more friendly interface for straightforward logistic regression.
Dropped the coefficient convergence plots and simplified mnlm output.
New for mnlm: prediction in probability space, response fit plots, and a summary function.
New for pls: predict, summary, and print functions.  Better plot and output labeling.
topic modeling remains in beta release.

1.4-1 (23 March 2011)
---
Fixed a potential color bug in plot.topics by setting explicit breaks.

1.4 (21 March 2011)
---
'topics' material lacks the companion paper, and is still in beta release.
Added time-check to allow for escape from running c-code.
IDF in tfidf now divides by number of containing docs plus a very tiny number.
New predict function for 'mnlm' objects
Updated example for congress109 now uses common-scores.

1.3-1 (17 March 2011)
---
Make the mnlm initialization more robust for huge 'p'.
Added null category for mnlm to remove identification from the user.
Improved binomial example for mnlm.
mnlm now returns 'loadings' and 'intercept', instead of raw coefficients matrix

1.3 (10 March 2011)
---
Functionality and documentation for topic modeling
Added the wsj-ibm data.

1.2-2 (23 February 2011)
---
Changes to corr, sdev, and tfidf to take better advantage of slam functionality

1.2-1 (22 February 2011)
---
Fix interaction with slam to account for package update.

1.2 (12 February 2011)
---
Improved handling of sparsity in large count matrices.
Add the document tf-idf function
Allow binning of V for faster inference with factors and large datasets.

1.1 (21 January 2011)
---
Minor changes: commented out we8there mnlm for faster vignette, and
	cleaned up the plot.pls function.

1.0 (20 January 2011)
---
Initial version: mnlm and pls for sentiment analysis
