https://sourceforge.net/p/infos/wiki/PeakAddSubSplitComb/Recent changes to PeakAddSubSplitCombenThu, 08 Sep 2016 14:44:10 -0000https://sourceforge.net/p/infos/wiki/PeakAddSubSplitComb/<div class="markdown_content"><pre>--- v1 +++ v2 @@ -1,4 +1,4 @@ ###Peak Addition, Removal, Splitting, and Combination If a fixed peak list is not used, then INFOS attempts to optimize the peak list, such that the spectrum is well fit, but fitting of noise is minimized. A cutoff is used to determine what peak height is no longer considered noise, and is used to add/remove/split peaks. If no user settings are given, then the FitSpec function will analyze the noise level, and the peak heights and attempt to set the cutoff for noise such that ~1% of peaks that are fitted are statistically likely to be noise. Furthermore, FitSpec will determine how well a peak can fit the noise in the spectrum, and use this information to decide when combining peaks reduces over-fitting of the spectrum. -These methods of evaluating the noise for fitting are powerful in obtaining optimal fits of spectra. However, the main limitation here is that they avoid over-fitting only noise. Therefore, artifacts above the noise level will be fit. Poor baselines are a severe problem in this case. A constant offset will cause mis-evaluation of noise, and uneven baselines will make defining a good noise cutoff impossible. Additional problems will arise if the selected lineshape type (‘gauss’, ‘lorentz’, ‘mixXX’) is not a good match for the experimental data, so that FitSpec will add extra peaks to fit out the mismatch. +These methods of evaluating the noise for fitting are powerful in obtaining optimal fits of spectra. However, the main limitation here is that they avoid over-fitting only noise. Therefore, artifacts above the noise level will be fit. Poor baselines are a severe problem in this case. A constant offset will cause mis-evaluation of noise, and uneven baselines will make defining a good noise cutoff impossible. Additional problems will arise if the selected lineshape type ([‘gauss’, ‘lorentz’, ‘mixXX’](/p/infos/wiki/dn_signal_decay)) is not a good match for the experimental data, so that FitSpec will add extra peaks to fit out the mismatch. </pre> </div>A. SmithThu, 08 Sep 2016 14:44:10 -0000https://sourceforge.net94b12cd47461133b57d062aac80db70c167ab776https://sourceforge.net/p/infos/wiki/PeakAddSubSplitComb/<div class="markdown_content"><h3 id="peak-addition-removal-splitting-and-combination">Peak Addition, Removal, Splitting, and Combination</h3> <p>If a fixed peak list is not used, then INFOS attempts to optimize the peak list, such that the spectrum is well fit, but fitting of noise is minimized. A cutoff is used to determine what peak height is no longer considered noise, and is used to add/remove/split peaks. If no user settings are given, then the FitSpec function will analyze the noise level, and the peak heights and attempt to set the cutoff for noise such that ~1% of peaks that are fitted are statistically likely to be noise. Furthermore, FitSpec will determine how well a peak can fit the noise in the spectrum, and use this information to decide when combining peaks reduces over-fitting of the spectrum.</p> <p>These methods of evaluating the noise for fitting are powerful in obtaining optimal fits of spectra. However, the main limitation here is that they avoid over-fitting only noise. Therefore, artifacts above the noise level will be fit. Poor baselines are a severe problem in this case. A constant offset will cause mis-evaluation of noise, and uneven baselines will make defining a good noise cutoff impossible. Additional problems will arise if the selected lineshape type (‘gauss’, ‘lorentz’, ‘mixXX’) is not a good match for the experimental data, so that FitSpec will add extra peaks to fit out the mismatch. </p></div>A. SmithThu, 08 Sep 2016 09:51:04 -0000https://sourceforge.net007e05dc9cb52459b707515b012467376bdda1cb