Analysis Service

Optional R and Python statistical analysis for pre- and post-visualization workflows

The Express server runs R or Python scripts for statistical analysis. Data stays under user control: the browser sends schema and rows to a self-hostable proxy; scripts read JSON on stdin and write JSON on stdout.

Source: server/analysis/ · runner: server/analysis/runner.js · endpoint: POST /api/analyze.

1 Requirements

1.1 R

install.packages(
  c("jsonlite", "dplyr", "tidyr", "broom", "car", "pwr", "lavaan"),
  repos = "https://cloud.r-project.org"
)

Scripts live under server/analysis/r/:

Script	Analysis
`descriptive.R`	Summary stats, optional group-by
`regression.R`	Linear regression / ANOVA
`power.R`	Power analysis
`mediation.R`	X → M → Y mediation
`factorial.R`	Factorial ANOVA (2–3 factors)

1.2 Python

pip install -r server/requirements.txt

Dependencies: pandas, scipy, statsmodels, numpy. Parallel scripts live under server/analysis/python/.

2 Environment

Variable	Description	Default
`ANALYSIS_TIMEOUT_MS`	Script timeout	`60000`
`R_PATH`	Path to `Rscript`	`Rscript`
`PYTHON_PATH`	Path to Python	`python3`

3 Endpoint contract

POST /api/analyze

Body:

{
  "engine": "r",
  "analysisType": "descriptive",
  "data": [{ "x": 1, "y": 2 }],
  "config": { "columns": ["x", "y"], "groupBy": null }
}

Field	Values
`engine`	`r` \| `python`
`analysisType`	`descriptive` \| `regression` \| `power` \| `mediation` \| `factorial`
`data`	Array of row objects (≤ 1,000 recommended)
`config`	Type-specific configuration

4 Example — descriptive (R)

# Excerpt from server/analysis/r/descriptive.R
suppressPackageStartupMessages({
  library(jsonlite)
  library(dplyr)
  library(tidyr)
})

input <- readLines(file("stdin"), warn = FALSE)
parsed <- fromJSON(paste(input, collapse = "\n"))
data <- as.data.frame(parsed$data)
config <- parsed$config

# Computes n, mean, sd, min, max, missing per numeric column
# Optional groupBy produces by_group summaries

5 UI integration

AnalysisPanel.jsx — engine selector, analysis type, config forms, Run
AnalysisResults.jsx — renders structured results
Hooks: useAnalysis.js · service: analysisService.js

Pre-viz analysis can inform the recommendation prompt; post-viz analysis supports follow-up statistical questions after a chart is chosen.

6 Privacy note

Analysis is opt-in. Recommendation calls never send full datasets — only schema + ≤5 sample rows. Prefer a self-hosted proxy in production so API keys and analysis data never leave your infrastructure.

--- title: "Analysis Service" subtitle: "Optional R and Python statistical analysis for pre- and post-visualization workflows" --- The Express server runs R or Python scripts for statistical analysis. Data stays under user control: the browser sends schema and rows to a self-hostable proxy; scripts read JSON on stdin and write JSON on stdout. Source: `server/analysis/` · runner: `server/analysis/runner.js` · endpoint: `POST /api/analyze`. ## Requirements ### R ```r install.packages( c("jsonlite", "dplyr", "tidyr", "broom", "car", "pwr", "lavaan"), repos = "https://cloud.r-project.org" ) ``` Scripts live under `server/analysis/r/`: | Script | Analysis | |--------|----------| | `descriptive.R` | Summary stats, optional group-by | | `regression.R` | Linear regression / ANOVA | | `power.R` | Power analysis | | `mediation.R` | X → M → Y mediation | | `factorial.R` | Factorial ANOVA (2–3 factors) | ### Python ```bash pip install -r server/requirements.txt ``` Dependencies: `pandas`, `scipy`, `statsmodels`, `numpy`. Parallel scripts live under `server/analysis/python/`. ## Environment | Variable | Description | Default | |----------|-------------|---------| | `ANALYSIS_TIMEOUT_MS` | Script timeout | `60000` | | `R_PATH` | Path to `Rscript` | `Rscript` | | `PYTHON_PATH` | Path to Python | `python3` | ## Endpoint contract `POST /api/analyze` Body: ```json { "engine": "r", "analysisType": "descriptive", "data": [{ "x": 1, "y": 2 }], "config": { "columns": ["x", "y"], "groupBy": null } } ``` | Field | Values | |-------|--------| | `engine` | `r` \| `python` | | `analysisType` | `descriptive` \| `regression` \| `power` \| `mediation` \| `factorial` | | `data` | Array of row objects (≤ 1,000 recommended) | | `config` | Type-specific configuration | ## Example — descriptive (R) ```r # Excerpt from server/analysis/r/descriptive.R suppressPackageStartupMessages({ library(jsonlite) library(dplyr) library(tidyr) }) input <- readLines(file("stdin"), warn = FALSE) parsed <- fromJSON(paste(input, collapse = "\n")) data <- as.data.frame(parsed$data) config <- parsed$config # Computes n, mean, sd, min, max, missing per numeric column # Optional groupBy produces by_group summaries ``` ## UI integration - `AnalysisPanel.jsx` — engine selector, analysis type, config forms, Run - `AnalysisResults.jsx` — renders structured results - Hooks: `useAnalysis.js` · service: `analysisService.js` Pre-viz analysis can inform the recommendation prompt; post-viz analysis supports follow-up statistical questions after a chart is chosen. ## Privacy note Analysis is **opt-in**. Recommendation calls never send full datasets — only schema + ≤5 sample rows. Prefer a self-hosted proxy in production so API keys and analysis data never leave your infrastructure.