created a small summary for October 2024

notes/open-notebook.2024.10.summary.md

@@ -0,0 +1,40 @@
+---
+id: a3aygyjlxham4yh44m6d0xr
+title: 2024-10-summary
+desc: ''
+updated: 1730464372633
+created: 1730448238478
+traitIds:
+  - open-notebook-mvisani
+---
+
+# Summary of October 2024
+
+## Start of the thesis
+
+## Random Markov Field
+We started the implementation of the random markov field. The main discussion
+was how can we store all the data- that we need in a reasonable amount of memory. To do so we decided
+to use a uint64 vecore to store the data where each bit represents a different value of the class. 
+The first 16 bits are used to store the counter of the MCMC, for counting how many times our combination
+(species-molecule) was a 1. The 17th bit is to store the actual state of the combination (if it is 1 or 0).
+The next 47 bits are used as index of the combination in the molecule and species vectors.
+
+We are now discussing if a vector is the best way to store the data or if we should use a hashmap
+(`std::unordered_map`).
+It seems like a vector with a "hinted" binary search could lead to extremely fast access to the data.
+
+### Z dimension
+For the Z dimension of the model, since we don't need to store the counter, we decided to go with a int32 where the 
+value is the coordinate of the vector and the sign is the state of the combination.
+
+## MS2 compound or compound to MS2
+Luca has been working on different models to predict the classification of a molecule. He has also 
+been working on the classification of MS2 spectra to chemical classifications. We would ideally try to develop
+a model that can predict a smiles or a chemical fingerprint given a MS2 spectrum.
+
+## RDKit Rust
+I am also working from time to time to create a binding beween C++ and Rust for the RDKit library. I am 
+a bit in contact with the creators of the crate but they are not very responsive. MAybe I should 
+fully develop my own fork.  
+