ML EPS REG: Difference between revisions

Latest revision as of 08:48, 16 May 2022

ML_EPS_REG = [real]
Default: ML_EPS_REG = 1E-15

Description: Initial value for the threshold of the eigenvalues of the covariance matrix in the evidence approximation.

This threshold is used to determine which eigenvalues $\lambda_{k}$ of the covariance matrix $\mathbf{\Phi}^{\mathrm{T}}\mathbf{\Phi}/\sigma^{2}_{\mathrm{v}}$ are used in the optimization of the regularization parameters $\sigma^{2}_{\mathrm{w}}$ and $\sigma^{2}_{\mathrm{v}}$ determined by the following equations

${\displaystyle \sigma^{2}_{\mathrm{w}}=\frac{|\mathbf{\bar{w}}|^{2}}{\gamma}, }$

${\displaystyle \sigma^{2}_{\mathrm{v}}=\frac{|\mathbf{T}-\mathbf{\phi}\mathbf{\bar{w}}|^{2}}{M-\gamma}, }$

${\displaystyle \gamma=\sum\limits_{k=1}^{N_{\mathrm{B}}} \frac{\lambda_{k}}{\lambda_{k}+1/\sigma^{2}_{\mathrm{w}}} }$ .

All eigenvalues satisfying $\lambda_{i} / \lambda_{\mathrm{max}}$ > ML_EPS_REG are included in the above equations, whereas smaller eigenvalues are disregarded (they are anyway potentially inaccurate because of loss of significance).

If at any point during iterating the above equations, the quadratic norm of errors (eight column of REGR/REGRF in ML_LOGFILE) becomes too large (more than 1.2 times larger than in previous iterations), the code assumes that numerical issues (loss of significance) have occurred, and then ML_EPS_REG is automatically doubled. Furthermore, if the regression does not converge within 10 steps, ML_EPS_REG is also increased by a factor of 4. The maximum allowed iteration depths is 50 (the iteration number is the second entry of REGR/REGRF in ML_LOGFILE). When 50 iterations are reached, no force field is created and there is most likely something seriously wrong in the calculation.

The seventh entry of REGR/REGRF in the ML_LOGFILE shows the ratio of the regularization ( $\sigma_{v}^{2}/ \sigma_{w}^{2}$ ) and the largest eigenvalue. Usually this number is a number with many varying digits. If this number becomes a "well rounded" number (e.g. 1.00000000E-14), this is an indication that the cap for the current ML_EPS_REG is reached. That means that regularization becomes crucial.

@@ Line 1: / Line 1: @@
-{{TAGDEF|ML_EPS_REG|[real]|1E-14}}
+{{DISPLAYTITLE:ML_EPS_REG}}
+{{TAGDEF|ML_EPS_REG|[real]|1E-15}}
 Description: Initial value for the threshold of the eigenvalues of the covariance matrix in the evidence approximation.
@@ Line 17: / Line 18: @@
 </math>.
-All eigenvalues satisfying <math>\lambda_{i} / \lambda_{\mathrm{max}} </math> > {{TAG|ML_EPS_REG}} are contributing by the above equations.
+All eigenvalues satisfying <math>\lambda_{i} / \lambda_{\mathrm{max}} </math> > {{TAG|ML_EPS_REG}} are included in the above equations, whereas smaller eigenvalues are disregarded (they are anyway potentially inaccurate because of loss of significance).
-The smaller the value of {{TAG|ML_EPS_REG}} the the smaller the error of the fit. But at the same time the effects of overfitting increase. We determined empirically that the default value of 1E-14 is a safe value in most cases. If a user desires higher accuracy of the force field, the user can try to set {{TAG|ML_EPS_REG}}=1E-15 or lower, but the user should very carefully monitor the convergence of the calculations as described below.
+If at any point during iterating the above equations, the quadratic norm of errors (eight column of <code>REGR/REGRF</code> in [[ML_LOGFILE]]) becomes too large (more than 1.2 times larger than in previous iterations), the code assumes that numerical issues
+(loss of significance) have occurred, and  then {{TAG|ML_EPS_REG}} is automatically doubled. Furthermore, if the regression does not converge within 10 steps, {{TAG|ML_EPS_REG}} is also increased by a factor of 4. The maximum allowed iteration depths is 50 (the iteration number is the second entry of <code>REGR/REGRF</code> in [[ML_LOGFILE]]). When 50 iterations are reached, no force field is created and there is most likely something seriously wrong in the calculation.
-If at any point in the iterations of the Evidence Approximation  of the square of the quadratic norm of errors (eigth column of <code>REGR/REGRF</code> in [[ML_LOGFILE]]) gets too big (more than 1.2 times larger than before) then {{TAG|ML_EPS_REG}} is doubled. Whenever the regression doesn't converge within 10 steps {{TAG|ML_EPS_REG}} is increased by a factor of 4. The counter for the 10 strucures is reset whenever {{TAG|ML_EPS_REG}} is changed. The maximum allowed iteration steps are 50, but when this value is reached no force field is created and there is most likely something seriously wrong in the calculation.
+The seventh entry of <code>REGR/REGRF</code> in the [[ML_LOGFILE]] shows the ratio of the regularization (<math>\sigma_{v}^{2}/ \sigma_{w}^{2}</math>) and the largest eigenvalue. Usually this number is a number with many varying digits. If this number becomes a "well rounded" number (e.g. 1.00000000E-14), this is an indication that the cap for the current {{TAG|ML_EPS_REG}} is reached. That means that regularization becomes crucial.
-The seventh entry of <code>REGR/REGRF</code> in the [[ML_LOGFILE]] shows the ratio of the regularization (<math>\sigma_{v}^{2}/ \sigma_{w}^{2}</math>) and the largest eigenvalue. Usually this number is a number with many varying digits. If this number becomes a "well rounded" number (e.g. 1.00000000E-14), it is an indication that the cap for the current {{TAG|ML_EPS_REG}} is reached. That means that regularization becomes crucial.
+== Related tags and articles ==
-== Related Tags and Sections ==
 {{TAG|ML_LMLFF}}, {{TAG|ML_IALGO_LINREG}}, {{TAG|ML_IREG}}, {{TAG|ML_SIGV0}}, {{TAG|ML_SIGW0}}
@@ Line 33: / Line 32: @@
 ----
-[[Category:INCAR]][[Category:Machine Learning]][[Category:Machine Learned Force Fields]][[Category: Alpha]]
+[[Category:INCAR tag]][[Category:Machine-learned force fields]]