Last updated: 2020-04-16

Checks: 7 0

Knit directory: mr_mash_test/

This reproducible R Markdown analysis was created with workflowr (version 1.6.1). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.

R Markdown file: up-to-date

Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.

Environment: empty

Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.

Seed: set.seed(20200328)

The command set.seed(20200328) was run prior to running the code in the R Markdown file. Setting a seed ensures that any results that rely on randomness, e.g. subsampling or permutations, are reproducible.

Session information: recorded

Great job! Recording the operating system, R version, and package versions is critical for reproducibility.

Cache: none

Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.

File paths: relative

Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.

Repository version: f6f65c4

Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility.

The results in this page were generated with repository version f6f65c4. See the Past versions tab to see a history of the changes made to the R Markdown and HTML files.

Note that you need to be careful to ensure that all relevant files for the analysis have been committed to Git prior to generating the results (you can use wflow_publish or wflow_git_commit). workflowr only checks the R Markdown file, but you know if there are other scripts or data files that it depends on. Below is the status of the Git repository when the results were generated: 


Ignored files:
    Ignored:    .sos/
    Ignored:    code/fit_mr_mash.66662433.err
    Ignored:    code/fit_mr_mash.66662433.out
    Ignored:    dsc/.sos/
    Ignored:    dsc/outfiles/
    Ignored:    output/dsc.html
    Ignored:    output/dsc/

Untracked files:
    Untracked:  code/plot_test.R

Unstaged changes:
    Modified:   dsc/midway2.yml

Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.

These are the previous versions of the repository in which changes were made to the R Markdown (analysis/results_accuracy.Rmd) and HTML (docs/results_accuracy.html) files. If you’ve configured a remote Git repository (see ?wflow_git_remote), click on the hyperlinks in the table below to view the files as they were in that past version.

File Version Author Date Message
html 422658e fmorgante 2020-04-16 Build site.
html 04f5485 fmorgante 2020-03-30 Build site.
html cc50c05 fmorgante 2020-03-30 Build site.
Rmd ec880d7 fmorgante 2020-03-30 Adjust date format
html 378e278 fmorgante 2020-03-30 Build site.
Rmd e4b5d3f fmorgante 2020-03-30 Add simulation 4
html 40cd0cf fmorgante 2020-03-30 Build site.
html b48d34d fmorgante 2020-03-30 Build site.
Rmd 6ff19bc fmorgante 2020-03-30 Fix typo in simulation 3
html 7a2afe7 fmorgante 2020-03-30 Build site.
Rmd b3eb015 fmorgante 2020-03-30 Add simulation 3
html f7e2f68 fmorgante 2020-03-30 Build site.
Rmd 6063335 fmorgante 2020-03-30 Fix typo in simulation 2
html 4f5c291 fmorgante 2020-03-30 Build site.
Rmd f4eb72a fmorgante 2020-03-30 Add simulation 2
html 8621d7d fmorgante 2020-03-30 Build site.
Rmd 1cfb1f8 fmorgante 2020-03-30 Slight improvements
html d5773c9 fmorgante 2020-03-30 Build site.
html 07a9b4d fmorgante 2020-03-30 Build site.
Rmd f416bb4 fmorgante 2020-03-30 Tweak prediction results
html f2c716e fmorgante 2020-03-30 Build site.
Rmd f9a9283 fmorgante 2020-03-30 Tweak prediction results
html 8e7a5b5 fmorgante 2020-03-30 Build site.
Rmd cf442a2 fmorgante 2020-03-30 Fix prediction results
html 4e9aad6 fmorgante 2020-03-30 Build site.
Rmd 1f94def fmorgante 2020-03-30 Tweak prediction results
html 7911e81 fmorgante 2020-03-30 Build site.
Rmd a6fd418 fmorgante 2020-03-30 Add prediction results
html e1707b2 fmorgante 2020-03-30 Build site.
Rmd 2b58162 fmorgante 2020-03-30 Start adding prediction results
html f2451af fmorgante 2020-03-29 Build site.
html 2fe7214 fmorgante 2020-03-29 Build site.
Rmd d43e6a6 fmorgante 2020-03-29 Set current date automatically
html dbf0d8e fmorgante 2020-03-29 Build site.
html c71dfff fmorgante 2020-03-29 Build site.
Rmd bafd72f fmorgante 2020-03-29 Modify titles and author
html 796c93e fmorgante 2020-03-29 Build site.
Rmd ed9843a fmorgante 2020-03-29 Add additional pages 
options(stringsAsFactors = FALSE)

accuracy <- function(Y, Yhat) {
  bias <- rep(NA, ncol(Y))
  r2 <- rep(NA, ncol(Y))
  mse <- rep(NA, ncol(Y))
  
  for(i in 1:ncol(Y)){
    fit  <- lm(Y[, i] ~ Yhat[, i])
    bias[i] <- coef(fit)[2] 
    r2[i] <- summary(fit)$r.squared
    mse[i] <- mean((Y[, i] - Yhat[, i])^2)
  }
  
  return(list(bias=bias, r2=r2, mse=mse))
}

Simulation 1 – Shared effects, independent variables

dat1 <- readRDS("output/fit_mr_mash_n600_p1000_p_caus50_r5_pve0.5_sigmaoffdiag1_sigmascale0.8_gammaoffdiag0_gammascale0.8_Voffdiag0.2_Vscale0_updatew0TRUE_updatew0TRUE_updatew0methodmixsqp_updateVTRUE.rds")
n1 <- dat1$params$n
p1 <- dat1$params$p
p_causal1 <- dat1$params$p_causal
r1 <- dat1$params$r
k1 <- length(dat1$fit$w0)
pve1 <- dat1$params$pve
prop_testset1 <- dat1$params$prop_testset
B1 <- dat1$inputs$B
V1 <- dat1$inputs$V
Sigma1 <- dat1$inputs$Sigma
Gamma1 <- dat1$inputs$Gamma
Ytrain1 <- dat1$Ytrain
Ytest1 <- dat1$Ytest
mu11 <- dat1$fit$mu1
fitted1 <- dat1$fit$fitted
Yhat_test1 <- dat1$Yhat_test

The results below are based on simulation with 600 samples, 1000 variables of which 50 were causal, 5 responses with a per-response proportion of variance explained (PVE) of 0.5. Variables, X, were drawn from MVN(0, Gamma), causal effects, B, were drawn from MVN(0, Sigma). The responses, Y, were drawn from MN(XB, I, V).

cat("Gamma (First 5 elements)")
Gamma (First 5 elements)
Gamma1[1:5, 1:5]
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.0  0.0  0.0  0.0
[2,]  0.0  0.8  0.0  0.0  0.0
[3,]  0.0  0.0  0.8  0.0  0.0
[4,]  0.0  0.0  0.0  0.8  0.0
[5,]  0.0  0.0  0.0  0.0  0.8
cat("Sigma")
Sigma
Sigma1
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.8  0.8  0.8  0.8
[2,]  0.8  0.8  0.8  0.8  0.8
[3,]  0.8  0.8  0.8  0.8  0.8
[4,]  0.8  0.8  0.8  0.8  0.8
[5,]  0.8  0.8  0.8  0.8  0.8
cat("V")
V
V1
         [,1]     [,2]     [,3]     [,4]     [,5]
[1,] 25.55836  0.00000  0.00000  0.00000  0.00000
[2,]  0.00000 25.55836  0.00000  0.00000  0.00000
[3,]  0.00000  0.00000 25.55836  0.00000  0.00000
[4,]  0.00000  0.00000  0.00000 25.55836  0.00000
[5,]  0.00000  0.00000  0.00000  0.00000 25.55836

mr.mash was fitted to the training data (80% of the data) updating V and updating the prior weights using mixSQP. Then, responses were predicted on the test data (20% of the data). The mixture prior consisted of 101 components.

In the plots below, each color/symbol defines a diffrent response.

Here, we compare the estimated effects with the true effects.

plot(B1[, 1], mu11[, 1], xlab="True effects", ylab="Estimated effects", main="True vs Estimated Effects", pch=1, cex.lab=1.5, cex.axis=1.5)
points(B1[, 2], mu11[, 2], col="blue", pch=2)
points(B1[, 3], mu11[, 3], col="red", pch=3)
points(B1[, 4], mu11[, 4], col="green", pch=4)
points(B1[, 5], mu11[, 5], col="yellow", pch=8)

Version Author Date
422658e fmorgante 2020-04-16
8621d7d fmorgante 2020-03-30
4e9aad6 fmorgante 2020-03-30
7911e81 fmorgante 2020-03-30
e1707b2 fmorgante 2020-03-30

Then, we compare the predicted responses with the true responses in the training data (left panel) and test data (right panel).

par(mfrow=c(1,2))
plot(Ytrain1[, 1], fitted1[, 1], xlab="True responses", ylab="Fitted values", main="True vs Fitted values \nTraining data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytrain1[, 2], fitted1[, 2], col="blue", pch=2)
points(Ytrain1[, 3], fitted1[, 3], col="red", pch=3)
points(Ytrain1[, 4], fitted1[, 4], col="green", pch=4)
points(Ytrain1[, 5], fitted1[, 5], col="yellow", pch=8)
abline(0, 1)

plot(Ytrain1[, 1], fitted1[, 1], xlab="True responses", ylab="Predicted responses", main="True vs Predicted Responses \nTest data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytest1[, 2], Yhat_test1[, 2], col="blue", pch=2)
points(Ytest1[, 3], Yhat_test1[, 3], col="red", pch=3)
points(Ytest1[, 4], Yhat_test1[, 4], col="green", pch=4)
points(Ytest1[, 5], Yhat_test1[, 5], col="yellow", pch=8)
abline(0, 1)

Version Author Date
422658e fmorgante 2020-04-16
8621d7d fmorgante 2020-03-30
4e9aad6 fmorgante 2020-03-30
7911e81 fmorgante 2020-03-30 
par(mfrow=c(1,1))

r2_train1 <- round(accuracy(Ytrain1, fitted1)$r2, 4)
r2_test1 <- round(accuracy(Ytest1, Yhat_test1)$r2, 4)
bias_train1 <- round(accuracy(Ytrain1, fitted1)$bias, 4)
bias_test1 <- round(accuracy(Ytest1, Yhat_test1)$bias, 4)
mse_train1 <- round(accuracy(Ytrain1, fitted1)$mse, 4)
mse_test1 <- round(accuracy(Ytest1, Yhat_test1)$mse, 4)

acc1 <- rbind(r2_train1, r2_test1, bias_train1, bias_test1, mse_train1, mse_test1)
colnames(acc1) <- paste0("Y", seq(1, r1))
part_metric1 <- c("Training data r2", "Test data r2", "Training data bias", "Test data bias" , "Training data MSE" , "Test data MSE")
res1 <- data.frame(part_metric1, acc1)
colnames(res1)[1] <- c("Partition_metric")
rownames(res1) <- NULL
print(res1)
    Partition_metric      Y1      Y2      Y3      Y4      Y5
1   Training data r2  0.5338  0.5418  0.4904  0.5239  0.5620
2       Test data r2  0.4591  0.4476  0.4630  0.4523  0.4237
3 Training data bias  1.0606  1.0586  0.9935  1.0013  1.0971
4     Test data bias  0.9892  1.0660  1.0666  1.0905  0.9659
5  Training data MSE 24.6985 23.8540 25.3913 22.4603 23.9106
6      Test data MSE 24.5054 29.4890 27.1642 29.6734 26.7496

Simulation 2 – Independent effects, independent variables

dat2 <- readRDS("output/fit_mr_mash_n600_p1000_p_caus50_r5_pve0.5_sigmaoffdiag0_sigmascale0.8_gammaoffdiag0_gammascale0.8_Voffdiag0.2_Vscale0_updatew0TRUE_updatew0TRUE_updatew0methodmixsqp_updateVTRUE.rds")
n2 <- dat2$params$n
p2 <- dat2$params$p
p_causal2 <- dat2$params$p_causal
r2 <- dat2$params$r
k2 <- length(dat2$fit$w0)
pve2 <- dat2$params$pve
prop_testset2 <- dat2$params$prop_testset
B2 <- dat2$inputs$B
V2 <- dat2$inputs$V
Sigma2 <- dat2$inputs$Sigma
Gamma2 <- dat2$inputs$Gamma
Ytrain2 <- dat2$Ytrain
Ytest2 <- dat2$Ytest
mu12 <- dat2$fit$mu1
fitted2 <- dat2$fit$fitted
Yhat_test2 <- dat2$Yhat_test

The results below are based on simulation with 600 samples, 1000 variables of which 50 were causal, 5 responses with a per-response proportion of variance explained (PVE) of 0.5. Variables, X, were drawn from MVN(0, Gamma), causal effects, B, were drawn from MVN(0, Sigma). The responses, Y, were drawn from MN(XB, I, V).

cat("Gamma (First 5 elements)")
Gamma (First 5 elements)
Gamma2[1:5, 1:5]
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.0  0.0  0.0  0.0
[2,]  0.0  0.8  0.0  0.0  0.0
[3,]  0.0  0.0  0.8  0.0  0.0
[4,]  0.0  0.0  0.0  0.8  0.0
[5,]  0.0  0.0  0.0  0.0  0.8
cat("Sigma")
Sigma
Sigma2
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.0  0.0  0.0  0.0
[2,]  0.0  0.8  0.0  0.0  0.0
[3,]  0.0  0.0  0.8  0.0  0.0
[4,]  0.0  0.0  0.0  0.8  0.0
[5,]  0.0  0.0  0.0  0.0  0.8
cat("V")
V
V2
         [,1]     [,2]     [,3]     [,4]     [,5]
[1,] 39.87305  0.00000  0.00000  0.00000  0.00000
[2,]  0.00000 24.41042  0.00000  0.00000  0.00000
[3,]  0.00000  0.00000 27.69452  0.00000  0.00000
[4,]  0.00000  0.00000  0.00000 25.53166  0.00000
[5,]  0.00000  0.00000  0.00000  0.00000 29.00472

mr.mash was fitted to the training data (80% of the data) updating V and updating the prior weights using mixSQP. Then, responses were predicted on the test data (20% of the data). The mixture prior consisted of 101 components.

In the plots below, each color/symbol defines a diffrent response.

Here, we compare the estimated effects with the true effects.

plot(B2[, 1], mu12[, 1], xlab="True effects", ylab="Estimated effects", main="True vs Estimated Effects", pch=1, cex.lab=1.5, cex.axis=1.5)
points(B2[, 2], mu12[, 2], col="blue", pch=2)
points(B2[, 3], mu12[, 3], col="red", pch=3)
points(B2[, 4], mu12[, 4], col="green", pch=4)
points(B2[, 5], mu12[, 5], col="yellow", pch=8)

Version Author Date
422658e fmorgante 2020-04-16
4f5c291 fmorgante 2020-03-30

Then, we compare the predicted responses with the true responses in the training data (left panel) and test data (right panel).

par(mfrow=c(1,2))
plot(Ytrain2[, 1], fitted2[, 1], xlab="True responses", ylab="Fitted values", main="True vs Fitted values \nTraining data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytrain2[, 2], fitted2[, 2], col="blue", pch=2)
points(Ytrain2[, 3], fitted2[, 3], col="red", pch=3)
points(Ytrain2[, 4], fitted2[, 4], col="green", pch=4)
points(Ytrain2[, 5], fitted2[, 5], col="yellow", pch=8)
abline(0, 1)

plot(Ytrain2[, 1], fitted2[, 1], xlab="True responses", ylab="Predicted responses", main="True vs Predicted Responses \nTest data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytest2[, 2], Yhat_test2[, 2], col="blue", pch=2)
points(Ytest2[, 3], Yhat_test2[, 3], col="red", pch=3)
points(Ytest2[, 4], Yhat_test2[, 4], col="green", pch=4)
points(Ytest2[, 5], Yhat_test2[, 5], col="yellow", pch=8)
abline(0, 1)

Version Author Date
422658e fmorgante 2020-04-16
4f5c291 fmorgante 2020-03-30 
par(mfrow=c(1,1))

r2_train2 <- round(accuracy(Ytrain2, fitted2)$r2, 4)
r2_test2 <- round(accuracy(Ytest2, Yhat_test2)$r2, 4)
bias_train2 <- round(accuracy(Ytrain2, fitted2)$bias, 4)
bias_test2 <- round(accuracy(Ytest2, Yhat_test2)$bias, 4)
mse_train2 <- round(accuracy(Ytrain2, fitted2)$mse, 4)
mse_test2 <- round(accuracy(Ytest2, Yhat_test2)$mse, 4)

acc2 <- rbind(r2_train2, r2_test2, bias_train2, bias_test2, mse_train2, mse_test2)
colnames(acc2) <- paste0("Y", seq(1, r2))
part_metric2 <- c("Training data r2", "Test data r2", "Training data bias", "Test data bias" , "Training data MSE" , "Test data MSE")
res2 <- data.frame(part_metric2, acc2)
colnames(res2)[1] <- c("Partition_metric")
rownames(res2) <- NULL
print(res2)
    Partition_metric      Y1      Y2      Y3      Y4      Y5
1   Training data r2  0.5373  0.5341  0.5146  0.5716  0.5570
2       Test data r2  0.4600  0.3854  0.4395  0.4473  0.4948
3 Training data bias  1.1737  1.1069  1.1264  1.1019  1.1270
4     Test data bias  0.9847  0.9370  1.1124  1.1859  0.9449
5  Training data MSE 38.1690 22.6568 25.8972 21.0610 26.7705
6      Test data MSE 41.8959 28.1329 34.3800 34.3338 30.6942

Simulation 3 – Shared effects, correlated variables

dat3 <- readRDS("output/fit_mr_mash_n600_p1000_p_caus50_r5_pve0.5_sigmaoffdiag1_sigmascale0.8_gammaoffdiag0.5_gammascale0.8_Voffdiag0.2_Vscale0_updatew0TRUE_updatew0TRUE_updatew0methodmixsqp_updateVTRUE.rds")
n3 <- dat3$params$n
p3 <- dat3$params$p
p_causal3 <- dat3$params$p_causal
r3 <- dat3$params$r
k3 <- length(dat3$fit$w0)
pve3 <- dat3$params$pve
prop_testset3 <- dat3$params$prop_testset
B3 <- dat3$inputs$B
V3 <- dat3$inputs$V
Sigma3 <- dat3$inputs$Sigma
Gamma3 <- dat3$inputs$Gamma
Ytrain3 <- dat3$Ytrain
Ytest3 <- dat3$Ytest
mu13 <- dat3$fit$mu1
fitted3 <- dat3$fit$fitted
Yhat_test3 <- dat3$Yhat_test

The results below are based on simulation with 600 samples, 1000 variables of which 50 were causal, 5 responses with a per-response proportion of variance explained (PVE) of 0.5. Variables, X, were drawn from MVN(0, Gamma), causal effects, B, were drawn from MVN(0, Sigma). The responses, Y, were drawn from MN(XB, I, V).

cat("Gamma (First 5 elements)")
Gamma (First 5 elements)
Gamma3[1:5, 1:5]
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.4  0.4  0.4  0.4
[2,]  0.4  0.8  0.4  0.4  0.4
[3,]  0.4  0.4  0.8  0.4  0.4
[4,]  0.4  0.4  0.4  0.8  0.4
[5,]  0.4  0.4  0.4  0.4  0.8
cat("Sigma")
Sigma
Sigma3
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.8  0.8  0.8  0.8
[2,]  0.8  0.8  0.8  0.8  0.8
[3,]  0.8  0.8  0.8  0.8  0.8
[4,]  0.8  0.8  0.8  0.8  0.8
[5,]  0.8  0.8  0.8  0.8  0.8
cat("V")
V
V3
         [,1]     [,2]     [,3]     [,4]     [,5]
[1,] 13.98626  0.00000  0.00000  0.00000  0.00000
[2,]  0.00000 13.98625  0.00000  0.00000  0.00000
[3,]  0.00000  0.00000 13.98625  0.00000  0.00000
[4,]  0.00000  0.00000  0.00000 13.98625  0.00000
[5,]  0.00000  0.00000  0.00000  0.00000 13.98625

mr.mash was fitted to the training data (80% of the data) updating V and updating the prior weights using mixSQP. Then, responses were predicted on the test data (20% of the data). The mixture prior consisted of 101 components.

In the plots below, each color/symbol defines a diffrent response.

Here, we compare the estimated effects with the true effects.

plot(B3[, 1], mu13[, 1], xlab="True effects", ylab="Estimated effects", main="True vs Estimated Effects", pch=1, cex.lab=1.5, cex.axis=1.5)
points(B3[, 2], mu13[, 2], col="blue", pch=2)
points(B3[, 3], mu13[, 3], col="red", pch=3)
points(B3[, 4], mu13[, 4], col="green", pch=4)
points(B3[, 5], mu13[, 5], col="yellow", pch=8)

Version Author Date
422658e fmorgante 2020-04-16
b48d34d fmorgante 2020-03-30
7a2afe7 fmorgante 2020-03-30

Then, we compare the predicted responses with the true responses in the training data (left panel) and test data (right panel).

par(mfrow=c(1,2))
plot(Ytrain3[, 1], fitted3[, 1], xlab="True responses", ylab="Fitted values", main="True vs Fitted values \nTraining data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytrain3[, 2], fitted3[, 2], col="blue", pch=2)
points(Ytrain3[, 3], fitted3[, 3], col="red", pch=3)
points(Ytrain3[, 4], fitted3[, 4], col="green", pch=4)
points(Ytrain3[, 5], fitted3[, 5], col="yellow", pch=8)
abline(0, 1)

plot(Ytrain3[, 1], fitted3[, 1], xlab="True responses", ylab="Predicted responses", main="True vs Predicted Responses \nTest data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytest3[, 2], Yhat_test3[, 2], col="blue", pch=2)
points(Ytest3[, 3], Yhat_test3[, 3], col="red", pch=3)
points(Ytest3[, 4], Yhat_test3[, 4], col="green", pch=4)
points(Ytest3[, 5], Yhat_test3[, 5], col="yellow", pch=8)
abline(0, 1)

Version Author Date
422658e fmorgante 2020-04-16
7a2afe7 fmorgante 2020-03-30 
par(mfrow=c(1,1))

r2_train3 <- round(accuracy(Ytrain3, fitted3)$r2, 4)
r2_test3 <- round(accuracy(Ytest3, Yhat_test3)$r2, 4)
bias_train3 <- round(accuracy(Ytrain3, fitted3)$bias, 4)
bias_test3 <- round(accuracy(Ytest3, Yhat_test3)$bias, 4)
mse_train3 <- round(accuracy(Ytrain3, fitted3)$mse, 4)
mse_test3 <- round(accuracy(Ytest3, Yhat_test3)$mse, 4)

acc3 <- rbind(r2_train3, r2_test3, bias_train3, bias_test3, mse_train3, mse_test3)
colnames(acc3) <- paste0("Y", seq(1, r3))
part_metric3 <- c("Training data r2", "Test data r2", "Training data bias", "Test data bias" , "Training data MSE" , "Test data MSE")
res3 <- data.frame(part_metric3, acc3)
colnames(res3)[1] <- c("Partition_metric")
rownames(res3) <- NULL
print(res3)
    Partition_metric      Y1      Y2      Y3      Y4      Y5
1   Training data r2  0.4892  0.5037  0.4710  0.4907  0.5390
2       Test data r2  0.4148  0.4341  0.4220  0.4639  0.4251
3 Training data bias  1.0358  1.0376  0.9876  0.9817  1.0887
4     Test data bias  1.0015  1.0979  1.0552  1.1736  1.0079
5  Training data MSE 14.4128 14.0238 14.5096 12.8917 13.6560
6      Test data MSE 14.0750 15.9244 15.3882 15.9627 14.0499

Simulation 4 – Independent effects, correlated variables

dat4 <- readRDS("output/fit_mr_mash_n600_p1000_p_caus50_r5_pve0.5_sigmaoffdiag0_sigmascale0.8_gammaoffdiag0.5_gammascale0.8_Voffdiag0.2_Vscale0_updatew0TRUE_updatew0TRUE_updatew0methodmixsqp_updateVTRUE.rds")
n4 <- dat4$params$n
p4 <- dat4$params$p
p_causal4 <- dat4$params$p_causal
r4 <- dat4$params$r
k4 <- length(dat4$fit$w0)
pve4 <- dat4$params$pve
prop_testset4 <- dat4$params$prop_testset
B4 <- dat4$inputs$B
V4 <- dat4$inputs$V
Sigma4 <- dat4$inputs$Sigma
Gamma4 <- dat4$inputs$Gamma
Ytrain4 <- dat4$Ytrain
Ytest4 <- dat4$Ytest
mu14 <- dat4$fit$mu1
fitted4 <- dat4$fit$fitted
Yhat_test4 <- dat4$Yhat_test

The results below are based on simulation with 600 samples, 1000 variables of which 50 were causal, 5 responses with a per-response proportion of variance explained (PVE) of 0.5. Variables, X, were drawn from MVN(0, Gamma), causal effects, B, were drawn from MVN(0, Sigma). The responses, Y, were drawn from MN(XB, I, V).

cat("Gamma (First 5 elements)")
Gamma (First 5 elements)
Gamma4[1:5, 1:5]
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.4  0.4  0.4  0.4
[2,]  0.4  0.8  0.4  0.4  0.4
[3,]  0.4  0.4  0.8  0.4  0.4
[4,]  0.4  0.4  0.4  0.8  0.4
[5,]  0.4  0.4  0.4  0.4  0.8
cat("Sigma")
Sigma
Sigma4
     [,1] [,2] [,3] [,4] [,5]
[1,]  0.8  0.0  0.0  0.0  0.0
[2,]  0.0  0.8  0.0  0.0  0.0
[3,]  0.0  0.0  0.8  0.0  0.0
[4,]  0.0  0.0  0.0  0.8  0.0
[5,]  0.0  0.0  0.0  0.0  0.8
cat("V")
V
V4
         [,1]     [,2]     [,3]     [,4]     [,5]
[1,] 31.75545  0.00000  0.00000  0.00000  0.00000
[2,]  0.00000 31.47091  0.00000  0.00000  0.00000
[3,]  0.00000  0.00000 14.55202  0.00000  0.00000
[4,]  0.00000  0.00000  0.00000 42.12604  0.00000
[5,]  0.00000  0.00000  0.00000  0.00000 15.37456

mr.mash was fitted to the training data (80% of the data) updating V and updating the prior weights using mixSQP. Then, responses were predicted on the test data (20% of the data). The mixture prior consisted of 101 components.

In the plots below, each color/symbol defines a diffrent response.

Here, we compare the estimated effects with the true effects.

plot(B4[, 1], mu14[, 1], xlab="True effects", ylab="Estimated effects", main="True vs Estimated Effects", pch=1, cex.lab=1.5, cex.axis=1.5)
points(B4[, 2], mu14[, 2], col="blue", pch=2)
points(B4[, 3], mu14[, 3], col="red", pch=3)
points(B4[, 4], mu14[, 4], col="green", pch=4)
points(B4[, 5], mu14[, 5], col="yellow", pch=8)

Version Author Date
422658e fmorgante 2020-04-16
378e278 fmorgante 2020-03-30

Then, we compare the predicted responses with the true responses in the training data (left panel) and test data (right panel).

par(mfrow=c(1,2))
plot(Ytrain4[, 1], fitted4[, 1], xlab="True responses", ylab="Fitted values", main="True vs Fitted values \nTraining data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytrain4[, 2], fitted4[, 2], col="blue", pch=2)
points(Ytrain4[, 3], fitted4[, 3], col="red", pch=3)
points(Ytrain4[, 4], fitted4[, 4], col="green", pch=4)
points(Ytrain4[, 5], fitted4[, 5], col="yellow", pch=8)
abline(0, 1)

plot(Ytrain4[, 1], fitted4[, 1], xlab="True responses", ylab="Predicted responses", main="True vs Predicted Responses \nTest data", pch=1, cex.lab=1.5, cex.axis=1.5)
points(Ytest4[, 2], Yhat_test4[, 2], col="blue", pch=2)
points(Ytest4[, 3], Yhat_test4[, 3], col="red", pch=3)
points(Ytest4[, 4], Yhat_test4[, 4], col="green", pch=4)
points(Ytest4[, 5], Yhat_test4[, 5], col="yellow", pch=8)
abline(0, 1)

Version Author Date
422658e fmorgante 2020-04-16
378e278 fmorgante 2020-03-30 
par(mfrow=c(1,1))

r2_train4 <- round(accuracy(Ytrain4, fitted4)$r2, 4)
r2_test4 <- round(accuracy(Ytest4, Yhat_test4)$r2, 4)
bias_train4 <- round(accuracy(Ytrain4, fitted4)$bias, 4)
bias_test4 <- round(accuracy(Ytest4, Yhat_test4)$bias, 4)
mse_train4 <- round(accuracy(Ytrain4, fitted4)$mse, 4)
mse_test4 <- round(accuracy(Ytest4, Yhat_test4)$mse, 4)

acc4 <- rbind(r2_train4, r2_test4, bias_train4, bias_test4, mse_train4, mse_test4)
colnames(acc4) <- paste0("Y", seq(1, r4))
part_metric4 <- c("Training data r2", "Test data r2", "Training data bias", "Test data bias" , "Training data MSE" , "Test data MSE")
res4 <- data.frame(part_metric4, acc4)
colnames(res4)[1] <- c("Partition_metric")
rownames(res4) <- NULL
print(res4)
    Partition_metric      Y1      Y2      Y3      Y4      Y5
1   Training data r2  0.4107  0.4811  0.3574  0.5330  0.4272
2       Test data r2  0.3990  0.2959  0.3766  0.3298  0.3542
3 Training data bias  1.1022  1.0512  1.0676  1.0356  1.0803
4     Test data bias  1.0478  0.7843  1.1576  0.9455  0.9463
5  Training data MSE 36.2095 32.5375 17.8351 39.5427 18.2162
6      Test data MSE 36.5641 38.1858 20.4419 62.0936 19.4614

sessionInfo()
R version 3.5.1 (2018-07-02)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Scientific Linux 7.4 (Nitrogen)

Matrix products: default
BLAS/LAPACK: /software/openblas-0.2.19-el7-x86_64/lib/libopenblas_haswellp-r0.2.19.so

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] workflowr_1.6.1 Rcpp_1.0.3      digest_0.6.25   later_0.7.5    
 [5] rprojroot_1.3-2 R6_2.4.1        backports_1.1.5 git2r_0.26.1   
 [9] magrittr_1.5    evaluate_0.12   stringi_1.4.3   fs_1.3.1       
[13] promises_1.0.1  whisker_0.3-2   rmarkdown_1.10  tools_3.5.1    
[17] stringr_1.4.0   glue_1.4.0      httpuv_1.4.5    yaml_2.2.1     
[21] compiler_3.5.1  htmltools_0.3.6 knitr_1.20