Mass-gap-review.md

@@ -57,3 +57,29 @@ This outputs (p_NS, p_EMB, p_massgap)
 | ----------- | ------- | --------------------- | ------------- |
 | M. Coughlin | ongoing | ligo.em-bright v1.1.0 |        :x:    |
 | S. Antier   | ongoing | ligo.em-bright v1.1.0 |        :x:    |
+
+##Comments
+
+- The code is running both from Coughlin and Antier condor https://git.ligo.org/emfollow/em-properties/em-bright/-/wikis/Mass-gap-review
+- clf_kwargs = {'n_estimators':50, 'criterion':'gini', 'max_depth': 20, 'max_features': None, 'min_samples_leaf': 5, 'min_samples_split': 20} is consistent with "https://git.ligo.org/emfollow/em-properties/mass_gap/-/blob/main/mass_gap_grid_search.ipynb" parameters grid search
+- Installation is very easy, thanks
+- Lack of understanding of conf.ini :
+--> executables : two times the same time, can you clarify ?
+--> lack of understanding and comments in config_ini parameters
+--> What is the definition of the snr here ? 
+--> Threshold of cfar : does it correspond to the FAR threshold for public alerts in O4 ? does it follow LL group definition ?
+--> "weignts": distance : how the distance is calculated here by the NN ?
+
+- Be careful on personal repositories present in the code
+--> to run the classifier output_dir -i /home/sushant.sharma-chaudhary/O2-HL-rates-injections
+or https://git.ligo.org/emfollow/em-properties/em-bright/-/merge_requests/41/diffs#0064bf8c2663d62fc40433c07c51bf36875714d1_162_192
+lin 49 PACKAGE_DATA_LINKS["MASS_GAP.pickle"] = 'https://git.ligo.org/sushant.sharma-chaudhary/em-bright-gp/-/raw/massgap/ligo/em_bright/data/MASS_GAP.pickle'
+
+- in the command em_bright_dag_writer
+--> -- mass-gap option, what does it refers to ?
+--> Layer   em_bright_extract       39        1      --> where 39 comes to ? Arbitrary ?
+
+- 120 m1 threshold discussed in https://git.ligo.org/emfollow/em-properties/mass_gap/-/blob/main/mass_gap_grid_search.ipynb
+--> What is the value to set-up this threshold so high while astronomers are interesting for prompt observation in EM_bright objets and with mass-gap ? so 1< m2 < 3 ?
+--> One suggestion would be to set a status as unknown when the rapid estimates clearly are deficient to avoid values that does not make sense.
+--> For mid terme, one useful study would be to train the NN only with this interesting sub-set of injections ( 1< m2 < 3)