Fit progress
Fit progress
This software's choice — A decision made by this software rather than by the field's sources; the explanation says why it was made.
While a fit runs, the line over the charts and the R-factor tab show how far it has got.
The line above the chart says what is being minimised and by which engine, the R-factor reached so far - in the form chosen under View > R-factor Scale - and how far it has fallen since the start. The status bar under the chart shows the elapsed time and the R-factor, in the same form, and says to use Stop to end the fit.
How far it has fallen is counted in orders of magnitude: "down 1.00" means ten times lower, "down 2.00" a hundred times lower. A percentage would stop near 100 % while the fit was still getting ten times better, and look as if it had stopped.
A long fit spends most of its time improving by less than a hundredth of an order of magnitude a minute, too little for "down 0.30" to change. Then the line adds what the last minute gained - "still improving, 0.00060 more in the last minute" - or says "no improvement in the last minute", so a fit that is still working does not look stopped. The fit goes on until the engine finds no worthwhile improvement left, which on a large model can take a long time; Stop ends it at once and keeps the model it reached.
Once the engine reports, the R-factor tab of the chart area draws the R-factor against elapsed time, on a logarithmic scale unless you choose another under View > R-factor Scale, because a fit improves by factors of ten: on a linear scale everything after the first moments would be a flat line along the bottom. The curve stays there when the fit ends, until the next one starts. The window does not switch tabs by itself: click R-factor to watch it.
Until the first report the R-factor tab is empty and the Model tab shows the data. An engine that reports no intermediate progress is said to be one, and its result appears when it finishes.
The value shown is the same R-factor the status bar reports while the fit runs. When the fit is over, the status bar shows the reduced chi-squared and R2 of the result instead.
Limitations
- The engine records a point each time it finds a better fit, not on every evaluation, so a long flat stretch means no better fit was found there rather than that nothing happened.
See also: R-factor scale
Animation Mode
This software's choice — A decision made by this software rather than by the field's sources; the explanation says why it was made.
With View > Animation Mode ticked, a running fit redraws the model over the data in the Model tab as it improves; the R-factor tab draws its progress either way.
The curves and the computed profile are redrawn about twice a second, each as it stood at the last better fit the engine found. The tables beside the chart keep the last finished result until the fit is over.
Animating makes a fit slower: the engine rebuilds the whole model for every frame it sends. The setting is remembered between sessions.
Limitations
- A frame shows the model at the last better fit the engine found, not at every evaluation, so the curves move in steps rather than smoothly.
R-factor scale
This software's choice — A decision made by this software rather than by the field's sources; the explanation says why it was made.
View > R-factor Scale chooses how the R-factor of a running fit is drawn on the progress chart, and how the status bar and the History tab give it; right-clicking that chart offers the same choices.
View > R-factor Scale > Automatic draws it on a logarithmic scale, because a fit improves by factors of ten: on a linear scale everything after the first moments would be a flat line along the bottom. View > R-factor Scale > Logarithmic chooses that scale outright, so it stays even if what Automatic means changes. Either way the marks on the axis are R-factor values.
View > R-factor Scale > Linear draws the R-factor as it is. It shows the first, large improvements well and hides the later, small ones.
View > R-factor Scale > Custom R-factor... draws it through a formula of your own, of x, the R-factor, and asks for the formula back from the drawn value, as a custom axis of the data does. log(x), the base-10 logarithm, with 10^x back draws the logarithm itself as a number.
The line above the chart gives the R-factor in the same form: "R-factor 1.32312E-6" on the linear scale, "log10 R-factor -5.87840" on the logarithmic one, and the value of your formula under the name you gave it on a custom one. The logarithm keeps moving visibly while a long fit improves by small steps. The status bar under the chart gives it in the same form, as "log10 R-factor: -5.87840", so one number never reads two ways on one screen.
The rows of the History tab give each model's R-factor in the same form, so models a power of ten apart are easy to tell apart: 5.46825E-7 and 1.74638E-5 look alike at a glance, while their logarithms, -6.26215 and -4.75786, do not. The details under the list always give the R-factor itself, the figure to quote.
The choice can be made while a fit runs, and the chart follows at its next update. It is saved with the project, like the axes of the data.
An analysis module may offer more ways of drawing it; each says in its own explanation what it does.
Limitations
- The scale changes only how the R-factor is drawn and written, never what the fit minimises or how it gets there, and never what a model records.