From c9fd69871165eb889be5421df518d8e35b2be027 Mon Sep 17 00:00:00 2001 From: vectorstorm Date: Tue, 16 Nov 2021 05:40:52 +1100 Subject: Reimplements WPM feature to be smaller & precise (#13902) * Reimplements WPM feature. - Now calculates exact WPM over the last up to three seconds of typing. - WPM_SMOOTHING removed, as it's no longer needed. - WPM_SAMPLE_SECONDS added, to specify how long a period to average WPM over, set to 5 seconds by default. - WPM_SAMPLE_PERIODS added, to specify how many sampling buffers we'll use. Each one uses one extra byte of space. Having more will lead to smoother decay of WPM values. Defaults to 50 (we're saving so many bytes of firmware space I felt like being extravagent, and this change is still a big size saving overall) - WPM_UNFILTERED option added (defaults to unset), which disables all filtering within the WPM feature. This saves some space in the firmware and also reduces latency between typing and the WPM calculation measuring it. (saves 70 bytes in my tests) - WPM_LAUNCH_CONTROL added (defaults to unset). When typing begins while the current displayed WPM value is zero, the WPM calculation only considers the time elapsed since typing began, not the whole WPM_SAMPLE_SECONDS buffer. The result of this is that the displayed WPM value much more rapidly reaches an accurate WPM value, even when results are being filtered. (costs 22 bytes in my tests) - Updates documentation to reflect changed options. Saves about 900 bytes, in my tests, compared against the previous implementation, with default settings. * Apply suggestions from code review Co-authored-by: Sergey Vlasov Co-authored-by: Trevor Powell Co-authored-by: Nick Brassel Co-authored-by: Sergey Vlasov --- docs/feature_wpm.md | 22 +++++++++++++++++----- 1 file changed, 17 insertions(+), 5 deletions(-) (limited to 'docs/feature_wpm.md') diff --git a/docs/feature_wpm.md b/docs/feature_wpm.md index 7e465e79b9..87145c97e3 100644 --- a/docs/feature_wpm.md +++ b/docs/feature_wpm.md @@ -10,11 +10,23 @@ For split keyboards using soft serial, the computed WPM score will be available ## Configuration -|Define |Default | Description | -|-----------------------------|--------------|------------------------------------------------------------------------------------------| -|`WPM_SMOOTHING` |`0.0487` | Sets the smoothing to about 40 keystrokes | -|`WPM_ESTIMATED_WORD_SIZE` |`5` | This is the value used when estimating average word size (for regression and normal use) | -|`WPM_ALLOW_COUNT_REGRESSOIN` |_Not defined_ | If defined allows the WPM to be decreased when hitting Delete or Backspace | +| Define | Default | Description | +|------------------------------|---------------|------------------------------------------------------------------------------------------| +| `WPM_ESTIMATED_WORD_SIZE` | `5` | This is the value used when estimating average word size (for regression and normal use) | +| `WPM_ALLOW_COUNT_REGRESSION` | _Not defined_ | If defined allows the WPM to be decreased when hitting Delete or Backspace | +| `WPM_UNFILTERED` | _Not defined_ | If undefined (the default), WPM values will be smoothed to avoid sudden changes in value | +| `WPM_SAMPLE_SECONDS` | `5` | This defines how many seconds of typing to average, when calculating WPM | +| `WPM_SAMPLE_PERIODS` | `50` | This defines how many sampling periods to use when calculating WPM | +| `WPM_LAUNCH_CONTROL` | _Not defined_ | If defined, WPM values will be calculated using partial buffers when typing begins | + +'WPM_UNFILTERED' is potentially useful if you're filtering data in some other way (and also because it reduces the code required for the WPM feature), or if reducing measurement latency to a minimum is important for you. + +Increasing 'WPM_SAMPLE_SECONDS' will give more smoothly changing WPM values at the expense of slightly more latency to the WPM calculation. + +Increasing 'WPM_SAMPLE_PERIODS' will improve the smoothness at which WPM decays once typing stops, at a cost of approximately this many bytes of firmware space. + +If 'WPM_LAUNCH_CONTROL' is defined, whenever WPM drops to zero, the next time typing begins WPM will be calculated based only on the time since that typing began, instead of the whole period of time specified by WPM_SAMPLE_SECONDS. This results in reaching an accurate WPM value much faster, even when filtering is enabled and a large WPM_SAMPLE_SECONDS value is specified. + ## Public Functions |Function |Description | -- cgit v1.2.3