1.6

* S3 methods predict.cgam is registered in 'NAMESPACE'

1.8

* Add p-value for each non-parametrically modeled component
* coneB uses columns of "delta" as edges

1.9
* Debug the p-value part in a summary table when there is a categorical covariate
* Debug predict.cgam when all predictors are unconstrained

1.10
* Update predict.cgam for binomial responses with a 95% confidence interval
* Default number of knots in cgam is increased by two
* Add anova table
* Add p-value for categorical predictors

1.11
* Add cgamm
* Add monotonic variance estimation

1.12
* Update ShapeSelect with user’s option to set an initial population size and mutation rate
* Update cgamm with prediction interval

1.13
* Stop initializing a vector for possible faces, which may cause memory leakage

1.15
* Add prediction for warped plane fit with visualization

1.16
* Add plotpersp in cgamm.R

1.17
* Change plotpersp to have methods for cgamm and csvy

1.18
* Add family = binomial in cgamm
* Update predict.cgam and predict.cgamm by using family$linkinv to get mu
* Revise cgam.pv argument: weights = wtkeep
* Revise xcoefs in cgamm
* Add ,... in persp functions
* Add ,... in cgamm for cgamm.control
* Change the following names to be consistent with csurvey: null_deviance = null.deviance, null_df = df.null, resid_df_obs = df.residual

1.19
* Update the default value of the space paranmeter as space = "Q" instead of space = "E" in each smooth symbolic function imposing shape or order constraint
* Update cgamm with a new paramter: nAGQ.

1.20
* Fix a bug in the predict.cgam function s.t. monotonicity of C.I. will be imposed when shapes = 9 or 10.
* Go back to the code
  :ch = muhat > 1e-5 & muhat < 1-(1e-5)
  :z[ch] = etahat[ch] + (y[ch]-muhat[ch])/muhat[ch]/(1-muhat[ch]) used in predict.cgam.
  
1.21
* Change the name of functions starting as ranef to start as .ranef: make these functions as hidden functions.

1.22
* Change plotpersp back to the version before csurvey was created.
* Eta in the binomial example in predict.cgam is changed to be 4*x^2 - 2
* llh.fun returns negative loglikelihood
* fix the bug in predict.cgam: sdhat = 1

1.23
* plotpersp needs to be compatible with csurvey

1.24
* add testpar 

1.25
* fix the formula0 bug in testpar
* makedmat_1D is faster
* newData is changed to be newdata in predict.cgam() to be consistent with predict.glm()

1.26
* fix a bug in makedelta when shape = 13,14,15,16: cannot scale x like this x_sc = (x - min(x)) / (max(x) - min(x))! -> NaN when predict for 1 data point

1.27
* fixed a bug in cgamm.fit: let coefs = coefskeep, not use coefs that include rf
* new plotpersp with plotpersp_control