added Week 1 materials

.gitignore

@@ -0,0 +1,3 @@
+*.o
+*.dll
+*.so

TMB examples/map argument/.Rapp.history

@@ -0,0 +1,36 @@
+# Simulate predictors#
+Factor = rep( 1:10, each=10)#
+Z = rnorm( length(unique(Factor)), mean=0, sd=1)#
+X0 = 0#
+#
+# Simulate response#
+Y = Z[Factor] + X0 + rnorm( length(Factor), mean=0, sd=1)
+library(TMB)#
+#
+# Compile model#
+Version = "linear_mixed_model"#
+compile( paste0(Version,".cpp") )#
+#
+# Build inputs#
+Data = list( "n_data"=length(Y), "n_factors"=length(unique(Factor)), "Factor"=Factor-1, "Y"=Y)#
+Parameters = list( "X0"=-10, "log_SD0"=2, "log_SDZ"=2, "Z"=rep(0,Data$n_factor) )
+Data
+Parameters
+# Turn off random effects#
+Map = list()#
+Map[["log_SDZ"]] = factor(NA)#
+Map[["Z"]] = factor( rep(NA,Data$n_factor) )
+Map
+# Build object#
+dyn.load( dynlib("linear_mixed_model") )#
+Obj = MakeADFun(data=Data, parameters=Parameters, map=Map)
+Obj
+Obj$par
+# Turn off random effects#
+Map = list()#
+Map[["log_SDZ"]] = factor(NA)
+Map
+# Build object#
+dyn.load( dynlib("linear_mixed_model") )#
+Obj = MakeADFun(data=Data, parameters=Parameters, map=Map)
+Obj$par

TMB examples/map argument/linear_mixed_model.R

@@ -0,0 +1,50 @@
+
+setwd( "C:/Users/James.Thorson/Desktop/Project_git/2016_Spatio-temporal_models/TMB examples/map argument" )
+
+######################
+# Simulate data
+######################
+
+# Simulate predictors
+Factor = rep( 1:10, each=10)
+Z = rnorm( length(unique(Factor)), mean=0, sd=1)
+X0 = 0
+
+# Simulate response
+Y = Z[Factor] + X0 + rnorm( length(Factor), mean=0, sd=1)
+
+######################
+# Run in TMB
+######################
+
+library(TMB)
+
+# Compile model
+Version = "linear_mixed_model"
+compile( paste0(Version,".cpp") )
+
+# Build inputs
+Data = list( "n_data"=length(Y), "n_factors"=length(unique(Factor)), "Factor"=Factor-1, "Y"=Y)
+Parameters = list( "X0"=-10, "log_SD0"=2, "log_SDZ"=2, "Z"=rep(0,Data$n_factor) )
+
+# Turn off random effects
+Map = list()
+Map[["log_SDZ"]] = factor(NA)
+Map[["Z"]] = factor( rep(NA,Data$n_factor) )
+
+# Build object
+dyn.load( dynlib("linear_mixed_model") )
+Obj = MakeADFun(data=Data, parameters=Parameters, map=Map)  #
+
+# Prove that function and gradient calls work
+Obj$fn( Obj$par )
+Obj$gr( Obj$par )
+
+# Optimize
+start_time = Sys.time()
+Opt = nlminb( start=Obj$par, objective=Obj$fn, gradient=Obj$gr, control=list("trace"=1) )
+
+# Get reporting and SEs
+Report = Obj$report()
+  SD = sdreport( Obj )
+

TMB examples/map argument/linear_mixed_model.cpp

@@ -0,0 +1,45 @@
+#include <TMB.hpp>
+template<class Type>
+Type objective_function<Type>::operator() ()
+{
+  // Data
+  DATA_INTEGER( n_data );
+  DATA_INTEGER( n_factors );
+  DATA_IVECTOR( Factor );
+  DATA_VECTOR( Y );                
+
+  // Parameters
+  PARAMETER( X0 );
+  PARAMETER( log_SD0 );
+  PARAMETER( log_SDZ );
+  PARAMETER_VECTOR( Z );
+
+  // Objective funcction
+  Type jnll = 0;
+
+  // Probability of data conditional on fixed and random effect values
+  for( int i=0; i<n_data; i++){
+    jnll -= dnorm( Y(i), X0 + Z(Factor(i)), exp(log_SD0), true );
+  }
+
+  // Probability of random coefficients
+  for( int i=0; i<n_factors; i++){
+    jnll -= dnorm( Z(i), Type(0.0), exp(log_SDZ), true );
+  }
+
+  // Reporting
+  Type SDZ = exp(log_SDZ);
+  Type SD0 = exp(log_SD0);
+
+  REPORT( SDZ );
+  REPORT( SD0 );
+  REPORT( Z );
+  REPORT( X0 );
+
+  ADREPORT( SDZ );
+  ADREPORT( SD0 );
+  ADREPORT( Z );
+  ADREPORT( X0 );
+
+  return jnll;
+}

Week 1 -- Likelihoods and linear models/Lab 1/Lab_1_code_V1.R

@@ -0,0 +1,156 @@
+
+setwd( "C:/Users/James.Thorson/Desktop/Project_git/2018_FSH556/Week 1 -- Likelihoods and linear models/Lab 1" )
+library(TMB)
+
+###########
+# Nonlinear optimization
+###########
+
+library("animation")  # Requires v2.12 or higher, also need to download: http://www.imagemagick.org/script/binary-releases.php
+
+# Nonlinear function
+RosenbrookFn = function(Params, Write=TRUE){
+  Dev = (1-Params[1])^2 + 100*(Params[2]-Params[1]^2)^2
+  if(Write==TRUE){
+    if("Trace.txt" %in% list.files()){
+      write.table(matrix(c(Dev,Params),nrow=1), file="Trace.txt", append=TRUE,col.names=FALSE,row.names=FALSE)
+    }else{
+      write.table(matrix(c(Dev,Params),nrow=1), file="Trace.txt", append=FALSE,col.names=FALSE,row.names=FALSE)
+    }
+  }
+  return(Dev)
+}
+
+# Generate contour
+X = seq(-10,20,length.out=1e3)
+Y = seq(-10,20,length.out=1e3)
+Z = matrix(NA, ncol=length(X), nrow=length(Y))
+for(xI in 1:length(X)){
+for(yI in 1:length(Y)){
+  Z[xI,yI] = RosenbrookFn(c(X[xI],Y[yI]), Write=FALSE)
+}}
+
+# Rosenbrook function
+png( file="Rosenbrook.png", width=6, height=6, res=200, units="in")
+  par( mar=c(3,3,2,0) )
+  contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=2)^5+1)
+  points( x=1, y=1, cex=2, col="red", lwd=2)
+dev.off()
+
+# Plot path -- Nelder Mead
+if("Trace.txt" %in% list.files()) file.remove("Trace.txt")
+Opt = nlminb( start=c(4,4), objective=RosenbrookFn, Write=TRUE )
+  Trace = read.table("Trace.txt" )
+contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1)
+  lines(x=Trace[,2], y=Trace[,3],col="red")
+  ani.options(interval=0.10, nmax=1e4)
+  saveVideo(expr={for(i in 1:nrow(Trace)){contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1); lines(x=Trace[max(1,i-10):i,2], y=Trace[max(1,i-10):i,3],col="red",lwd=3); points(x=c(1,Trace[i,2]),y=c(1,Trace[i,3]), col=c("black","red"),cex=2,lwd=3)}}, video.name="Nelder-Mead.mp4", convert = 'gm convert', clean=TRUE)         #
+
+# Plot path -- BFGS
+if("Trace.txt" %in% list.files()) file.remove("Trace.txt")
+Opt = optim( par=c(4,4), fn=RosenbrookFn, Write=TRUE )
+  Trace = read.table("Trace.txt" )
+contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1)
+  lines(x=Trace[,2], y=Trace[,3],col="red")
+  ani.options(interval=0.10, nmax=1e4)
+  saveVideo(expr={for(i in 1:nrow(Trace)){contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1); lines(x=Trace[max(1,i-10):i,2], y=Trace[max(1,i-10):i,3],col="red",lwd=3); points(x=c(1,Trace[i,2]),y=c(1,Trace[i,3]), col=c("black","red"),cex=2,lwd=3)}}, video.name="BFGS.mp4", convert = 'gm convert', clean=TRUE)         #
+
+# Plot path -- TMB
+# Step 1 -- make and compile template file
+compile( "Rosenbrook.cpp" )
+
+# Step 2 -- build inputs and object
+dyn.load( dynlib("Rosenbrook") )
+Params = list( "Params"=c(4,4) )
+Data = list( "dummy"=0 )
+Obj = MakeADFun( data=Data, parameters=Params, DLL="Rosenbrook")
+
+# Redefine functions
+Obj$fn_trace <- function( Params, ...){
+  Dev = Obj$fn( Params, ... )
+  if("Trace.txt" %in% list.files()){
+    write.table(matrix(c(Dev,Params),nrow=1), file="Trace.txt", append=TRUE,col.names=FALSE,row.names=FALSE)
+  }else{
+    write.table(matrix(c(Dev,Params),nrow=1), file="Trace.txt", append=FALSE,col.names=FALSE,row.names=FALSE)
+  }
+  return( Dev )
+}
+
+# Step 3 -- test and optimize
+if("Trace.txt" %in% list.files()) file.remove("Trace.txt")
+Opt = nlminb( start=Obj$par, objective=Obj$fn_trace, gradient=Obj$gr )
+  Trace = read.table("Trace.txt" )
+contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1)
+  lines(x=Trace[,2], y=Trace[,3],col="red")
+  ani.options(interval=0.10, nmax=1e4)
+  saveVideo(expr={for(i in 1:nrow(Trace)){contour(x=X, y=Y, z=Z, levels=seq(0,1000,by=10)^3+1); lines(x=Trace[max(1,i-10):i,2], y=Trace[max(1,i-10):i,3],col="red",lwd=3); points(x=c(1,Trace[i,2]),y=c(1,Trace[i,3]), col=c("black","red"),cex=2,lwd=3)}}, video.name="TMB.mp4", convert = 'gm convert', clean=TRUE)         #
+
+###########
+# Delta-model for canary rockfish
+###########
+
+devtools::install_github("nwfsc-assess/geostatistical_delta-GLMM")
+library( SpatialDeltaGLMM )
+
+#
+data(WCGBTS_Canary_example)
+CPUE = WCGBTS_Canary_example$HAUL_WT_KG
+X = cbind( "Intercept"=rep(1,length(CPUE)) )
+
+# Step 1 -- make and compile template file
+compile( "delta_model_v1.cpp" )
+
+# Step 2 -- build inputs and object
+dyn.load( dynlib("delta_model_v1") )
+Params = list("b_j"=rep(0,ncol(X)), "theta_z"=c(0,0))
+Data = list( "y_i"=CPUE, "X_ij"=X )
+Obj = MakeADFun( data=Data, parameters=Params, DLL="delta_model_v1")
+
+# Step 3 -- test and optimize
+Obj$fn( Obj$par )
+Obj$gr( Obj$par )
+Opt = nlminb( start=Obj$par, objective=Obj$fn, gradient=Obj$gr )
+Opt$diagnostics = data.frame( "name"=names(Obj$par), "Est"=Opt$par, "final_gradient"=as.vector(Obj$gr(Opt$par)))
+Opt$par # estimated parameters
+SD = sdreport( Obj ) # standard errors
+
+# Extract stuff
+Report = Obj$report()
+
+# Visualize fit
+png( file="Canary_histogram--with_fit.png", width=4, height=4, res=200, units="in")
+  par( mar=c(3,3,2,0), mgp=c(2,0.5,0), tck=-0.02)
+  hist( log(1+CPUE), freq=FALSE, col=rgb(1,0,0,0.2) )
+  Sim_CPUE = (1-rbinom(1e5, size=1, prob=Report$zero_prob)) * rlnorm(1e5, meanlog=Report$linpred_i, sdlog=Report$logsd)
+  hist( log(1+Sim_CPUE), freq=FALSE, add=TRUE, col=rgb(0,0,1,0.2) )
+  legend( "topright", bty="n", legend=c("Observed","Predicted"), fill=c("red","blue"))
+dev.off()
+
+###########
+# Crossvalidation experiment using delta-model for canary rockfish
+###########
+
+# Step 0 -- make and compile template file
+compile( "delta_model_v2.cpp" )
+dyn.load( dynlib("delta_model_v2") )
+
+# Step 1 -- divide into partitions
+K = 10
+Partition_i = sample( x=1:K, size=length(CPUE), replace=TRUE )
+PredNLL_k = rep(NA, K)
+
+# Step 2 --Loop through partitions
+for(k in 1:K){
+  Params = list("b_j"=rep(0,ncol(X)), "theta_z"=c(0,0))
+  Data = list( "y_i"=CPUE, "X_ij"=X, predTF_i=ifelse(Partition_i==k,1,0) )
+  Obj = MakeADFun( data=Data, parameters=Params, DLL="delta_model_v2")
+  Opt = nlminb( start=Obj$par, objective=Obj$fn, gradient=Obj$gr )
+  Report = Obj$report()
+  PredNLL_k[k] = Report$pred_jnll
+}
+
+# log-Predictive probability per datum
+mean( PredNLL_k / table(Partition_i) )
+
+
+

Week 1 -- Likelihoods and linear models/Lab 1/Rosenbrook.cpp

@@ -0,0 +1,17 @@
+// Space time
+#include <TMB.hpp>
+template<class Type>
+Type objective_function<Type>::operator() ()
+{
+  // Data
+  DATA_SCALAR( dummy );
+
+  // Parameters
+  PARAMETER_VECTOR( Params );
+
+  // Objective funcction
+  Type jnll = pow(1-Params(0),2) + 100*pow(Params(1)-pow(Params(0),2),2);
+
+  // Reporting
+  return jnll;
+}

Week 1 -- Likelihoods and linear models/Lab 1/Trace.txt

@@ -0,0 +1,42 @@
+14409 4 4
+2427.00072099677 3.00771735492052 4.12399658170315
+405.567414727713 2.65858999238928 5.06107186402148
+18.85016210181 2.54167027391276 6.05421323332969
+2.84052160663671 2.66012128811671 7.04717313042224
+2.48500715651973 2.57408553778493 6.6344380201405
+2.23172687072318 2.49322944264862 6.22065702314602
+3.18801765974419 2.16815385655468 4.56585654912003
+2.00271594240845 2.41024401150956 5.79747459749452
+1.87488605284059 2.3256970972798 5.37460137008961
+1.77194653364814 2.28315403564269 5.17737169908532
+1.35689715514021 2.14013048534639 4.55628389870529
+1.09787003107786 2.03695670644768 4.13416238108617
+1.00957541887104 1.85503830706771 3.38839545098624
+0.568839706248912 1.74096609143522 3.04503736138855
+0.434842044388779 1.64595911534794 2.69592287792184
+0.407023062869276 1.56884793522114 2.43239871108632
+0.330273705858859 1.5552667427509 2.40403824807302
+0.281940582356929 1.52955414047999 2.33564614545925
+0.236492322508975 1.48383012387979 2.19685210823917
+0.191985011699856 1.39169261567786 1.91717115075547
+0.145941492546445 1.37338807097366 1.87811838277943
+0.119799804021527 1.34216179988057 1.79617804158212
+0.105536983836215 1.25267970672711 1.54878833077767
+0.0647292580087596 1.25422104341677 1.5720658407961
+0.0601678332615207 1.24528231446488 1.55051781730963
+0.0484329717987803 1.2194670526587 1.48546531429903
+0.0683766725485752 1.07912909721802 1.13959667554488
+0.0311203129994762 1.16671083829298 1.35544546335281
+0.0204533108352537 1.11149011793839 1.22645301459011
+0.00854578852597585 1.0884630049999 1.18206827278944
+0.00524697907085572 1.06741031796208 1.13671369635572
+0.00323015633269444 1.04285120562733 1.08380509919993
+0.000662638916923775 1.02572915358576 1.05203970044365
+0.000156811462375771 1.01008182154083 1.01952253229918
+1.18957012499119e-05 1.00216171569509 1.00405935384206
+2.97447697019197e-07 1.00053469810139 1.0010804271676
+9.18162300049358e-09 0.999928188914863 0.999850038862464
+3.69301837436315e-13 1.00000039517812 1.00000083652307
+3.99639202020881e-19 1.00000000039871 1.00000000084648
+2.36046904364757e-27 0.999999999999961 0.999999999999925
+2.36046904364757e-27 0.999999999999961 0.999999999999925