Report M48 max delta

[vovodroid]

Description

Add maximum delta from mean value in M48 report - i.e. std::max(mean - min, max - mean).

Precision is also reduced from six to three to save screen space, I guess 0.001 is good enough.
I didn't change string buffer size, as I per my understanding original 8 bytes buffer is too small for precision six (plus leading zero, dot, and trailing null). So for precision 3 six byte buffer needed. Do I miss something?

Benefits

It could help to discover case when one probe deviate too much.

[thisiskeithb]

Precision is also reduced from six to three to save screen space, I guess 0.001 is good enough.

This will still overflow standard screens, so it should be <20 characters.

[vovodroid]

it should be <20 characters.

Dev/delta .001/.001 is 19. Is it suitable? I don't think it will be more than 1 mm.))

[vlsi]

What do you think of showing range (max-min) rather than 'max delta'?
I guess range would be easier to understand without reading the documentation

[vovodroid]

What do you think of showing range (max-min) rather than 'max delta'?

Well, it was my first approach. But next I thought that what we need to know is how single probe could divert from real value.

Assuming we have following result sets:

0.1 0.2 0.3 0.4 0.5 - range is 0.4, max delta is 0.2
0.1 0.1 0.1 0.1 0.5 range is 0.4, but max delta is 0.32

So first set presents case with even measurements distribution, and second set is case when one probe gives bad result.
From y point of view we would like to recognize send case, indicating some problem with probe. And actually deviation report show range.

[ellensp]

You might want to look at why CI keeps failing,,, Some platforms are older and the code needs to run on them all.

remove the std::max use max

[vlsi]

Well, the deviation would be different for both cases.

The standard deviation itself does not report the range, but it shows "deviation from the mean"

0.1 0.1 0.1 0.1 0.5 yields sigma of 0.16, 95% coinf. interval of 0.14
0.1 0.2 0.3 0.4 0.5 yields sigma of 0.14, 95% coinf. interval of 0.124, 99% coinf. interval of 0.163, 99% coinf. interval of 0.184

So it reporting range=0.4, 99% c.i. = 0.184 would more-or-less clarify the summary.

Then, it might make sense to show the values (e.g. diff from a mean)

[vovodroid]

0.1 0.1 0.1 0.1 0.5 yields sigma of 0.16
0.1 0.2 0.3 0.4 0.5 yields sigma of 0.14

That's what I mean! In first case sigma is only slightly worse than in second, but expected error is 0.4 in first case vs 0.2 in second. So max delta is more informative to detect probing problems.

remove the std::max use max

It fails, that's why I used std::

Marlin\src\gcode\calibrate\M48.cpp:266:183: error: 'max' cannot be used as a function
  266 |       ui.status_printf(0, F(S_FMT ": %s, " S_FMT ":%s"), GET_TEXT(MSG_M48_DEVIATION), dtostrf(sigma, 2, 3, sigma_str), GET_TEXT(MSG_M48_MAX_DELTA), dtostrf(max(mean - min, max - mean) , 2, 3, delta_str));

[vlsi]

So max delta is more informative to detect probing problems

If you show both "range", and "max delta", then it is more-or-less fine.
However, If you show only "max delta", then it is worse than showing "range" alone.

I might be missing something. Could you please refer to an article suggesting to use max delta descriptive statistic?

[vovodroid]

If you show both "range", and "max delta", then it is more-or-less fine.

Unfortunately there is no enough room on screen, especially on standard screens.

Could you please refer to an article suggesting to use max delta descriptive statistic?

It's called "maximum absolute deviation from the mean", I guess.

[vlsi]

0.1 0.1 0.1 0.1 0.5 has max abs deviation of 0.32
0.1 0.5 0.1 0.5 0.1 has max abs deviation of 0.24

Does it really mean the second probe is better? I do not think so.
What is the use of "max abs deviation" in the M48 results?

[vovodroid]

Does it really mean the second probe is better?

Yes. Assuming in the first case real value is 0.1, we see that worst probe error is 0.4. And in second case real value seems to be 0.3, and worst error is 0.2. Default probing number is 10, so worst probe will less affect mean (real) value.

What is the use of "max abs deviation" in the M48 results?

It can indicate probe problem, when one of several probes has large errors. Obviously it could be better to show all available information, but we are limited by screen size.

[thinkyhead]

On narrow screens without staus message scrolling enabled the status message will be too long to read, so I'd recommend checking for that before attempting to display an over-long message.

[vovodroid]

Do we really need 6 digits for deviation? Any meaning for < 0.001?

[thinkyhead]

Do we really need 6 digits for deviation? Any meaning for < 0.001?

No harm if it fits on the screen. People who deal with deviations tend to like precision.