Structural Equation Modeling with Deep Neural Network and Machine Learning
SEMdeep train and validate a custom (or data-driven) structural equation model (SEM) using layer-wise deep neural networks (DNNs) or node-wise machine learning (ML) algorithms. SEMdeep comes with the following functionalities:
Automated ML or DNN model training based on SEM network structures.
Network plot representation as interpretation diagram.
Model performance evaluation through regression and classification metrics.
Model variable importance computation through Shapley (R2) values, Gradient (or Connection) weight approach and significance tests of network inputs.
SEMdeep uses the deep learning framework ‘torch’. The torch package is native to R, so it’s computationally efficient, as there is no need to install Python or any other API, and DNNs can be trained on CPU, GPU and MacOS GPUs. Before using ‘SEMdeep’ make sure that the current version of ‘torch’ is installed and running:
install.packages("torch")
library(torch)
install_torch(reinstall = TRUE)Only for windows (not Linux or Mac). Some Windows distributions don’t have the Visual C++ runtime pre-installed, download from Microsoft VC_redist.x86.exe (R32) or VC_redist.x86.exe (R64) and install it.
For GPU setup, or if you have problems installing torch package, check out the installation help from the torch developer.
Then, the latest stable version can be installed from CRAN:
install.packages("SEMdeep")The latest development version can be installed from GitHub:
# install.packages("devtools")
devtools::install_github("BarbaraTarantino/SEMdeep")The full list of SEMdeep functions with examples is available at our website HERE.