MATLAB and R Programming (ME-705) - Unit 4 Short Notes

4.1 Advanced Data Handling & Preprocessing

4.1.1 Complex Data Structures

[!TIP] Common Pitfall: In R, stringsAsFactors = FALSE (default in recent versions) prevents automatic conversion of strings to factors. In MATLAB, table variables can be accessed via dot notation (T.VarName).

4.1.2 Data Import/Export from Diverse Sources

4.1.3 Data Cleaning & Transformation Pipelines

• Handling Missing Data:

  • Imputation Strategies: Mean/median/mode filling, forward/backward fill (fillmissing in MATLAB, tidyr::fill in R), model-based imputation.

  • Removal: rmmissing (MATLAB), na.omit (R).

• Reshaping Data:

  • Wide to Long (Melt): MATLAB: stack. R: tidyr::pivot_longer().

  • Long to Wide (Pivot): MATLAB: unstack. R: tidyr::pivot_wider().

• Pipeline Operations:

  • MATLAB: Use varfun, rowfun for group-wise operations. Chain with dot notation or findgroups + splitapply.

  • R: dplyr + %>% pipe operator. Key verbs: filter(), select(), mutate(), group_by(), summarise().

[!TIP] Exam Focus: Know the difference between pivot_longer()/pivot_wider() (tidyr, modern) vs gather()/spread() (tidyr, deprecated). In MATLAB, unstack requires a grouping variable.

4.2 Advanced Statistical Modeling & Machine Learning

4.2.1 Supervised Learning

[!TIP] Key Concept: Regularization (Ridge/Lasso) adds penalty term $\lambda$ to loss function to prevent overfitting. Lasso (L1) can drive coefficients to zero (feature selection).

4.2.2 Unsupervised Learning

• Clustering:

  • k-Means: MATLAB: kmeans. R: stats::kmeans(). Sensitive to initial centroids & scale.

  • Hierarchical: MATLAB: linkage, cluster. R: hclust(), cutree().

  • DBSCAN: MATLAB: dbscan (Statistics Toolbox). R: dbscan::dbscan().

• Dimensionality Reduction:

  • PCA: MATLAB: pca. R: prcomp() (recommended) or princomp().

  • t-SNE: MATLAB: tsne. R: Rtsne::Rtsne() or uwot::umap (for UMAP).

4.2.3 Time Series Analysis & Forecasting

• Stationarity Tests: ADF test (adftest in MATLAB, tseries::adf.test() in R).

• ARIMA/SARIMA:

  • Model Form: ARIMA$(p,d,q)$: $$\displaystyle \phi(B)(1-B)^d y_t = \theta(B)\epsilon_t $$, where $B$ is backshift operator.

  • MATLAB: arima(p,d,q), estimate, forecast.

  • R: forecast package: auto.arima() (automatically selects p,d,q). fable/tsibble (tidyverts): modern framework.

[!TIP] Common Pitfall: auto.arima in R uses stepwise search by default; set stepwise=FALSE for exhaustive search. Always check residuals (checkresiduals() in forecast).

4.3 Simulation & Optimization Techniques

4.3.1 Monte Carlo Methods

• Core Idea: Use repeated random sampling to estimate numerical results.

• Random Number Generation:

  • MATLAB: rand, randn, randi (control with rng).

  • R: runif, rnorm, sample (control with set.seed).

• Variance Reduction: Antithetic variates, control variates, importance sampling.

• Application Example (Option Pricing):

$$S_T = S_0 \exp\left((r - \frac{\sigma^2}{2})T + \sigma\sqrt{T}Z\right)$$

where $Z \sim N(0,1)$. Payoff = $$\displaystyle \max(S_T - K, 0) $$. Discount and average.

4.3.2 Deterministic Optimization

• Linear Programming (LP): Minimize $$\displaystyle c^Tx $$ s.t. $Ax \leq b$, $x \geq 0$.

  • MATLAB: linprog(f, A, b).

  • R: lpSolve::lp(direction="min", objective.in=f, const.mat=A, const.dir="<=", const.rhs=b).

• Integer Programming (IP): Add integrality constraints on some $$\displaystyle x_i $$.

  • MATLAB: intlinprog(f, intcon, A, b, Aeq, beq, lb, ub).

  • R: ompr::MIPModel() + ROI plugin (more modern) or lpSolve.

4.3.3 Nonlinear & Multi-objective Optimization

• Nonlinear:

  • MATLAB: fmincon (constrained), fminunc (unconstrained).

  • R: nloptr::nloptr().

• Global/Heuristic:

  • Genetic Algorithm (GA): MATLAB: ga. R: GA::ga().

• Multi-objective:

  • Pareto Front: MATLAB: paretosearch. R: mco::nsga2().

4.4 Advanced Visualization & Reporting

4.4.1 MATLAB Graphics Deep Dive

• Customization: title, xlabel, ylabel, legend, grid on. Colormaps: colormap('jet'), parula.

• Interactive Figures:

  • datacursormode(gcf, 'on'): Data cursor.

  • uicontrol: Create buttons/sliders. Write callback functions that update plot data.

• Geographic Plots: Requires Mapping Toolbox. geobasemap, geoplot(lat,lon), geoscatter.

4.4.2 R Visualization Ecosystem

• ggplot2 Grammar: ggplot(data, aes(x,y)) + geom_point() + geom_line() + labs() + theme_minimal().

  • Layers: Geoms (geom_*), Stats (stat_*), Position (position_*).

  • Scales: scale_x_log10(), scale_color_manual().

  • Faceting: facet_wrap(~var) or facet_grid(row~col).

• Interactive:

  • plotly: ggplotly(p) converts ggplot to interactive.

  • shiny: Build reactive web apps. Core: ui (layout), server (logic), shinyApp(ui, server).

• Dashboards: flexdashboard (R Markdown based) or shinydashboard (Shiny UI functions).

4.5 Performance, Debugging & Code Management

4.5.1 Profiling & Acceleration

4.5.2 Debugging Strategies

• MATLAB:

  • Set breakpoints in Editor.

  • dbstop if error: Stops on error.

  • dbstack: Show call stack.

  • keyboard: Pauses execution, enters debug mode.

• R:

  • browser(): Insert in code to pause.

  • traceback(): After error, shows call stack.

  • debugonce(fun): Step through function once.

  • RStudio: "Debug" pane, breakpoints, environment inspection.

4.5.3 Code Management & Reproducibility

• Version Control: Git (conceptual). Use .gitignore for large data/compiled files.

• Packages/Libraries:

  • MATLAB: Create toolboxes. Add via matlab.addons.install. Use matlab.addons.toolbox.installToolbox.

  • R: Package development with devtools/usethis. Document with roxygen2. Use renv::init() for dependency snapshot (like pipenv/conda).

• Reproducibility: Set seeds (rng in MATLAB, set.seed in R). Use project-oriented workflows (RStudio Projects, MATLAB Projects).

4.6 Interoperability & Specialized Applications

4.6.1 Calling R from MATLAB & Vice Versa

• MATLAB → R: Use system command to run Rscript.

  
  system('Rscript myscript.R arg1 arg2');
  
  

  Exchange data via files (.csv, .RData, .mat).

• R → MATLAB:

  • R.matlab package (R): readMat("file.mat"), writeMat().

  • reticulate (R → Python): Can indirectly exchange if Python has MATLAB engine.

  • MATLAB Engine for Python: From R, use reticulate to call Python which uses MATLAB Engine API.

4.6.2 Domain-Specific Application Patterns

4.7 Deployment & Sharing of Results

4.7.1 Creating Standalone Applications

• MATLAB:

  • MATLAB Compiler (mcc): Package apps into standalone executables or shared libraries.

  • MATLAB Web App Server: Deploy apps as web apps (requires Compiler & Web App Server).

  • App Designer: Build GUI apps.

• R:

  • Shiny Apps: Deploy to shinyapps.io, RStudio Connect, or Shiny Server (open-source/pro).

  • APIs: plumber package. Decorate R functions with @* annotations to create REST API.

4.7.2 Generating Reports & Publications

• MATLAB:

  • Live Scripts (.mlx): Combine code, output, formatted text.

  • Publish: publish('script.m', 'pdf') or 'html'. Uses markup in comments (%% sections).

• R:

  • R Markdown (.Rmd): Code chunks ({r}), inline R (r expr). Knit to HTML/PDF/Word.

  • Quarto: Next-gen, supports R, Python, Julia. More publishing-focused.

  • Key Feature: Dynamic report generation—code runs, results (tables, plots) embedded automatically.

[!TIP] Critical Difference: R Markdown/Quarto separates code and narrative clearly in a single file. MATLAB Live Scripts are more integrated but less portable. For reproducible research, R Markdown/Quarto is industry standard.