You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, converting a model object (e.g. StatMech) to a Nasa object is done in the following steps:
Use the model object to generate CpoR, HoRT, and SoR data.
Determine T_mid by iteratively partitioning a low range and a high range of temperatures and determining which T_mid value minimizes the fitting mean-squared error of CpoR.
Assign a values to each partition by curve fitting to data in (1)
While this exhaustive approach will always give an answer, it is presumably not the most efficient (with respect to time and finding a global minimum). We also make assumptions that the first minimum in error encountered is the global minimum, which is not necessarily true. This approach becomes more inefficient for Nasa9 objects that can have an arbitrary number of T_mid values.
We have attempted to use scipy's minimize functions where step 2 no longer exhaustive. Surprisingly, we find this method is ~50 times slower than the exhaustive method. This is likely due to the increased number of method calls for model.get_CpoR.
Since the error for the Nasa is not extremely sensitive to T_mid, we will favor the exhaustive method for now. However, I would still like to optimize this in the future.
The text was updated successfully, but these errors were encountered:
Currently, converting a model object (e.g.
StatMech
) to aNasa
object is done in the following steps:CpoR
,HoRT
, andSoR
data.T_mid
by iteratively partitioning a low range and a high range of temperatures and determining whichT_mid
value minimizes the fitting mean-squared error ofCpoR
.a
values to each partition by curve fitting to data in (1)While this exhaustive approach will always give an answer, it is presumably not the most efficient (with respect to time and finding a global minimum). We also make assumptions that the first minimum in error encountered is the global minimum, which is not necessarily true. This approach becomes more inefficient for
Nasa9
objects that can have an arbitrary number ofT_mid
values.We have attempted to use scipy's minimize functions where step 2 no longer exhaustive. Surprisingly, we find this method is ~50 times slower than the exhaustive method. This is likely due to the increased number of method calls for
model.get_CpoR
.Since the error for the
Nasa
is not extremely sensitive toT_mid
, we will favor the exhaustive method for now. However, I would still like to optimize this in the future.The text was updated successfully, but these errors were encountered: