fani-lab/OpeNTF

File Structure

• each training set will have it's own teamsvecs.pkl file that can be accessed through the following file path:
• the file structure is identical to that in the OpeNTF github repository

\---output
|	\---{domain}                            # e.g. dblp
|		\---{training dataset}              # e.g. toy.dblp.v12.json
|			teamsvecs.pkl

Teamsvecs Structure

Teamsvecs.pkl

{
    'skill': {spare matrix of n_teams x n_skills},
    'member': {sparse matrix of n_teams x n_members},
    'loc': {sparse matrix of n_teams x n_locs}                   // optional
}

skill

• every team in the entire dataset is represented as a row
• every skill in the entire dataset is represented as a column
• each row represents the skills that everyone on the team possesses

member

• every team in the entire dataset is represented as a row
• every member in the entire dataset is represented as a column
• each row represents all the members for a the given team

loc

• every geolocation in the entire dataset is represented as a row
• every skill in the entire dataset is represented as a column
• each row represents all the geolocations for a given team
• NOTE: will be set to None if the dataset does not include geolocation

Full Sparse Matrix

• the three sparse matrices from above can be aligned side by side to form a sparse matrix of n_teams x (n_skills + n_members + n_locs)

Example

• team 1 includes members m1 and m2, skills of s2 and s4, and has the geolocation of l2.
• team 2 includes members m2 and m3, skills of s1 and s2, and has the geolocation of l1.

s1	s2	s3	s4	m1	m2	m3	l1	l2
0	1	0	1	1	1	0	0	1
1	1	0	0	0	1	1	1	0

Links

• OpeNTF GitHub Repository