NMachine | small pack of machine learning algorithms | Machine Learning library

I am trying to read performance reports from multiple machines and would like to parse them and combine them to easily compare machine performances on single plots. Once divided into multiple csv I plan on reading them using pd.read_csv() and combine mutliple tools into single df's.

But in order to that I must first deal with & split up rather ugly csv files with semicolon separators.
The structure of the CSV is like this:

KEYWORD_01;;;...;;
COL_01;COL_02;COL03;...;COL_n;
Line_1;
Line_2;
Line_3;
...
Line_m;
KEYWORD_02;;;...;;
COL_01;COL_02;COL03;...;COL_x;
Line_1;
Line_2;
Line_3;
...
Line_y;
KEYWORD_03;;;...;;
COL_01;COL_02;COL03;...;COL_f;
Line_1;
Line_2;
Line_3;
...
Line_g;

Data csv file available here

The csv report is made of multiple sections each beginning with a fixed keyword (or keyphrase), each section has a set number of column (which can vary from section to section) and a dynamic number of rows depending on the number of events reported (cfr structure above).

1. I create a list with all my keywords called tpm_sections

   ...