Column feature LP Age

[benjadb]

Hello,
I don't understand the meaning of the LP Age feature and how is calculated. I'll appreciate it if someone could explain it to me.

Thank you

[benjadb]

Please, someone than can help me understand this feature

[gasse]

Hi @benjadb ,

I'm not sure which feature your refer to. Are you interested in the scaled_age feature from the NodeBipartite observation function ? You can find how this feature is calculated in Ecole, for both LP columns and rows, here:

ecole/libecole/src/observation/nodebipartite.cpp

Line 141 in e9522ef

out[idx(ColumnFeatures::scaled_age)] = static_cast<value_type>(SCIPcolGetAge(col)) / (n_lps + cste);

ecole/libecole/src/observation/nodebipartite.cpp

Line 241 in e9522ef

out[idx(RowFeatures::scaled_age)] = static_cast<value_type>(SCIProwGetAge(row)) / (n_lps + cste);

You can also look at SCIP's API for details about what the LP age means:
https://www.scipopt.org/doc/html/group__PublicRowMethods.php#gaf33c0582cf3c27fb12d93efdfb9b9156
https://www.scipopt.org/doc/html/group__PublicColumnMethods.php#ga417a303f33362d817f10602803511966

There it says:

gets the age of a column, i.e., the total number of successive times a column was in the LP and was 0.0 in the solution

Hope that helps.

Best,
Maxime

[benjadb]

I understood perfectly. Thank you!