Advise on the best compute tier for a tidylearn fit — tl_compute

Estimates runtime and feasibility on local CPU, local GPU (when available and applicable), and cloud GPU (stubbed until cloud integration lands), then returns a structured recommendation. Useful before kicking off a long fit — call this first to see whether the problem is laptop-sized, GPU-sized, or cloud-sized.

Usage

tl_compute_advisor(x, ...)

# S3 method for class 'character'
tl_compute_advisor(
  x,
  data,
  formula = NULL,
  hyperparams = list(),
  gpu_check = NULL,
  ...
)

# S3 method for class 'tidylearn_supervised'
tl_compute_advisor(
  x,
  data = NULL,
  formula = NULL,
  hyperparams = list(),
  gpu_check = NULL,
  ...
)

# Default S3 method
tl_compute_advisor(x, ...)

Arguments

x: Either a method name (character scalar — same names as accepted by tl_model(), supervised methods only) or a fitted tidylearn_supervised model. When given a fitted model, the advisor introspects its method and formula and advises on what a refit (or a fit on similar new data) would cost.
...: Unused, reserved for method-specific extensions.
data: A data frame. Required when x is a method name; optional when x is a fitted model (defaults to the model's training data).
formula: Optional formula. Used to determine the number of effective predictors. Ignored when x is a fitted model.
hyperparams: Named list of hyperparameters that affect runtime (e.g. list(nrounds = 1000) for xgboost, list(epochs = 50, units = 256) for deep learning). Missing entries fall back to per-method defaults.
gpu_check: Optional tidylearn_gpu_check object. If omitted, tl_check_gpu() is called once internally.

Value

An object of class tidylearn_compute_advice containing problem, local_cpu, local_gpu, cloud, recommendation, and reasoning. A print() method is provided.

Details

Estimates are deliberately rough — order-of-magnitude, not bills. Per-method scaling constants are calibrated against typical hardware and will be off by 2-3x in either direction for any individual job. Treat the recommendation as a starting point, not gospel.

Examples

# Estimating from a method name needs neither a GPU nor the backend
# package -- it is arithmetic over the problem dimensions
advice <- tl_compute_advisor("xgboost", iris, Species ~ .,
                             hyperparams = list(nrounds = 1000))
advice$recommendation
#> [1] "cpu"
print(advice)
#> <tidylearn compute advice>
#> Problem:        xgboost on 150 rows x 4 cols (~0.0 MB)
#> 
#> Tier estimates (order-of-magnitude):
#>   Local CPU:    0.0s   (peak RAM ~0 MB, 4 cores)
#>   Local GPU:    --   [not applicable]
#>   Cloud:        45.0s   (~$0.01) [T4 (16 GB VRAM / 16 GB RAM)]   [not configured]
#> 
#> Recommendation: cpu
#> 
#> Reasoning:
#>   - Estimated local CPU runtime ~0.0s. Cloud cold-start (~45s) would dominate; just run it locally.
#> 
#> Notes:
#>   - Method 'xgboost' could use GPU, but no GPU-capable backend was detected. See ?tl_check_gpu.
#>   - Cloud integration is not yet configured in tidylearn. Estimates shown so users can see the tier's shape; actual submission is not yet supported.

# \donttest{
# Dispatching on a fitted model requires the backend to be installed
if (requireNamespace("xgboost", quietly = TRUE)) {
  model <- tl_model(iris, Species ~ ., method = "xgboost")
  tl_compute_advisor(model)
}
#> <tidylearn compute advice>
#> Problem:        xgboost on 150 rows x 4 cols (~0.0 MB)
#> 
#> Tier estimates (order-of-magnitude):
#>   Local CPU:    0.0s   (peak RAM ~0 MB, 4 cores)
#>   Local GPU:    --   [not applicable]
#>   Cloud:        45.0s   (~$0.01) [T4 (16 GB VRAM / 16 GB RAM)]   [not configured]
#> 
#> Recommendation: cpu
#> 
#> Reasoning:
#>   - Estimated local CPU runtime ~0.0s. Cloud cold-start (~45s) would dominate; just run it locally.
#> 
#> Notes:
#>   - Method 'xgboost' could use GPU, but no GPU-capable backend was detected. See ?tl_check_gpu.
#>   - Cloud integration is not yet configured in tidylearn. Estimates shown so users can see the tier's shape; actual submission is not yet supported.
# }