Skip to contents

What's new in lactater 1.0

Source: vignettes/articles/whats-new.Rmd
whats-new.Rmd

lactater 1.0 is a different package from 0.3.0: every method follows its original paper and is solved exactly, the results come with checks and confidence intervals, and the thresholds go on to training zones, training load, comparisons between tests, race predictions and a report for the athlete. It is also much faster. The numbers of a test can change a little, for reasons given at the end.

What is new

  • Methods as their papers define them. Log-log as Beaver et al. (1985) drew it, the Dmax family with the line where each paper put it, LTP1 and LTP2 as the exact optimum of three segments, LTratio as an exact minimum, OBLA and Bsln+ solved on the curve itself; new IAT (Dickhuth) and Stegmann thresholds.
  • Plots that explain themselves. Each method drawn on the curve with its construction, a forest-style overview of every threshold with a table of values, zones colored by intensity domain, night and journal styles.
  • How sure, and what to double-check. Bootstrap intervals (ci = TRUE), lactate_checks() (including whether the LT1 and LT2 methods agree), lactate_diagnostics() and lactate_fit_quality().
  • From thresholds to training. lactate_zones() (three or five zones), training_load() (time in zone and the polarization index), predict_performance() and calibrate_performance().
  • From test to test. compare_tests() against the test-retest noise of each method, lactate_reliability() for your own noise, curve_shift(), curve_shape(), lactate_trend(), relative_thresholds() and reference_percentiles().
  • For the athlete and the lab. test_report() (a web page and a two-page PDF), lactate_report() for papers, a web form that records a test and read_recording() to read it back.
  • Beyond the step test. lactate_mlss() from constant-load bouts and vlamax() from a sprint test.
  • Data and inputs. lactate_tests (16 real and simulated tests across four sports), rowing, RPE, flexible targets (fixed, baseline_plus, iat_rise), one turn point for submaximal tests (ltp_breakpoints = 1), and .by for many tests in one call.

Faster

Every method is now computed in closed form or by a deterministic search, with no iterative model fitting, and eleven dependencies were dropped. On demo_data with every method, lactate_threshold() takes about 10 ms without plots where 0.3.0 took about 500 ms, some fifty times faster; with the plot of every method it takes 0.5 s instead of 0.9 s. A whole squad goes through .by in under a second, and bootstrap intervals with 500 replicates take a few seconds (measured on a laptop; your numbers will differ, the ratio less so).

Why some numbers change, and an example

Because the methods are now computed as their authors defined them, a threshold can land a little away from where 0.3.0 put it. The size of the change depends on the test: on a smooth curve OBLA and Bsln+ move by a watt or two, while a Log-log split or an LTP turn point can move more when the stages leave the fit a choice. Here is one example, cycling_female from lactate_tests, run with include_baseline = TRUE as the 0.3.0 values were computed (the default of 1.0 is FALSE). It is one test, not a rule: your own tests will differ by other amounts.

library(lactater)
library(dplyr)

## 0.3.0 run on the same test (fit = "3rd degree polynomial", include_baseline = TRUE)
old <- tibble(
  method = c("Log-log", "OBLA 2.0", "OBLA 2.5", "OBLA 3.0", "OBLA 3.5", "OBLA 4.0",
             "Bsln + 0.5", "Bsln + 1.0", "Bsln + 1.5", "Dmax", "ModDmax", "Exp-Dmax",
             "Log-Poly-ModDmax", "Log-Exp-ModDmax", "LTP1", "LTP2", "LTratio"),
  version_0.3.0 = c(199.1, 237.7, 245.9, 253.1, 260.2, 265.0, 223.1, 235.3, 244.4,
                    232.0, 254.5, 241.9, 254.1, 260.7, 203.4, 259.9, 190.8)
)
test <- lactate_tests |> filter(test == "cycling_female")
new <- lactate_threshold(test, intensity, lactate, include_baseline = TRUE, plot = FALSE)
## the unrounded intensities, to compare at 0.1 W
unrounded <- attr(new, "lactater")$intensity
old |>
  inner_join(tibble(method = names(unrounded), version_1.0 = round(unname(unrounded), 1)), by = "method") |>
  mutate(difference = version_1.0 - version_0.3.0) |>
  knitr::kable()
method version_0.3.0 version_1.0 difference
Log-log 199.1 199.3 0.2
OBLA 2.0 237.7 238.2 0.5
OBLA 2.5 245.9 246.9 1.0
OBLA 3.0 253.1 254.0 0.9
OBLA 3.5 260.2 260.2 0.0
OBLA 4.0 265.0 265.8 0.8
Bsln + 0.5 223.1 224.1 1.0
Bsln + 1.0 235.3 236.2 0.9
Bsln + 1.5 244.4 245.3 0.9
Dmax 232.0 231.2 -0.8
ModDmax 254.5 254.5 0.0
Exp-Dmax 241.9 241.9 0.0
Log-Poly-ModDmax 254.1 254.3 0.2
Log-Exp-ModDmax 260.7 260.0 -0.7
LTP1 203.4 207.5 4.1
LTP2 259.9 264.7 4.8
LTratio 190.8 190.8 0.0

What moved the numbers:

  • OBLA and Bsln+. The intensity at a target lactate is solved on the fitted curve itself (issue #8, thanks to Simon Dierickx).
  • Dmax family. Each variant takes the point where the fitted curve lies furthest below its line, from its start to the last stage, with the ends where each paper put them (dmax_anchor); the Log-log start is computed without the resting sample.
  • Log-log. Two straight lines that share the stage between them (Beaver et al., 1985), split at the stage with the lowest residual sum of squares.
  • LTP. The exact least-squares optimum on the measured stages, found by exhaustive search, instead of an iterative fit that depended on its starting values.
  • LTratio. The exact minimum, within the exercise stages.
  • Heart rate. Interpolated between the measured stages, so it follows the bend of the heart-rate curve.

See the news for the full list of changes.