Recipes & Canonical Workflows

This page serves as a compact, authoritative recipe guide for data scientists and automated coding agents using rtichoke in R.

---

1. Single Binary Classification Model

Evaluate predictions from one model on a single dataset.

library(rtichoke)

# Input vectors
probs_vector <- example_dat$estimated_probabilities
reals_vector <- example_dat$outcome

# ROC Curve
create_roc_curve(
  probs = list(probs_vector),
  reals = list(reals_vector)
)

# Precision-Recall Curve
create_precision_recall_curve(
  probs = list(probs_vector),
  reals = list(reals_vector)
)

# Decision Curve
create_decision_curve(
  probs = list(probs_vector),
  reals = list(reals_vector)
)

# Calibration Curve
create_calibration_curve(
  probs = list(probs_vector),
  reals = list(reals_vector)
)

---

2. Comparing Multiple Models on the Same Population

Compare multiple models evaluated on identical subjects and outcomes using a named list.

# Named list of predictions, single outcome vector in list
create_roc_curve(
  probs = list(
    "Model A (Good)" = example_dat$estimated_probabilities,
    "Model B (Bad)"  = example_dat$bad_model,
    "Model C (Random)" = example_dat$random_guess
  ),
  reals = list(example_dat$outcome)
)

---

3. Evaluating Across Distinct Subgroups or Split Sets

Compare one model across distinct cohorts (e.g., Train vs. Test split or demographic groups).

train_df <- example_dat[example_dat$type_of_set == "train", ]
test_df  <- example_dat[example_dat$type_of_set == "test", ]

create_roc_curve(
  probs = list(
    "Train Cohort" = train_df$estimated_probabilities,
    "Test Cohort"  = test_df$estimated_probabilities
  ),
  reals = list(
    "Train Cohort" = train_df$outcome,
    "Test Cohort"  = test_df$outcome
  )
)

---

4. Reusable Prepared Data Pipeline

Prepare metrics once and generate multiple visualizations efficiently.

perf_data <- prepare_performance_data(
  probs = list(
    "Model A" = example_dat$estimated_probabilities,
    "Model B" = example_dat$bad_model
  ),
  reals = list(example_dat$outcome)
)

# Reuse prepared object across all evaluation functions
plot_roc_curve(perf_data)
plot_precision_recall_curve(perf_data)
plot_gains_curve(perf_data)
plot_lift_curve(perf_data)
plot_decision_curve(perf_data)
render_performance_table(perf_data)

---

5. Automated HTML Summary Report

Generate a single interactive HTML file containing all evaluations and performance tables.

create_summary_report(
  probs = list("Primary Model" = example_dat$estimated_probabilities),
  reals = list(example_dat$outcome),
  file_path = "model_evaluation_report.html"
)

---

6. Supported Capabilities & Boundaries Summary

Feature / Output	Supported in rtichoke R	Input Format
ROC Curve	Yes	probs = list(...), reals = list(...)
Precision-Recall Curve	Yes	probs = list(...), reals = list(...)
Gains Curve	Yes	probs = list(...), reals = list(...)
Lift Curve	Yes	probs = list(...), reals = list(...)
Decision Curve (Net Benefit)	Yes	probs = list(...), reals = list(...)
Interventions Avoided	Yes	Integrated in Decision Curve outputs
Calibration Curve	Yes	Deciles / Smooth options available
Performance Tables	Yes	Threshold-specific metrics summary
HTML Summary Report	Yes	create_summary_report()
Time-Dependent Horizons (_times)	No	Not implemented in R package
Time-Dependent Calibration	No	Not implemented in R package