Basic Skills Inference #26
Conversation
hodsonjames
left a comment
hodsonjames
left a comment
There was a problem hiding this comment.
Just a few quick fixes. I'd like you to run a test over a few hundred profiles and look at what is being found, to get an idea of whether it's working as expected. I'd also like you to run an evaluation metric for overlap with the true set of raw_skills for profiles that do have a non-empty raw_skills field. That is, take n profiles that have 1 or more raw_skills, and see how many of those raw skills are found with this approach, per profile.
| keywords[v] = 1 | ||
|
|
||
| def infer_exp(self, p, e, return_scores = False): | ||
| if "role" not in e.keys() or e["role"] == None: |
There was a problem hiding this comment.
keys() yields a list, which loses you the computational advantage of a dictionary look-up. Instead, the better way to do this is either:
(a) if 'role' in e and e['role']:
(b) if e.get('role',None):
| v = set(items[2].replace('\n', '').lower().split(', ')) | ||
| self.vals[k] = v | ||
|
|
||
| def render_scores(self, data, scores, keywords): |
There was a problem hiding this comment.
Can you please add in-line comments to document the purpose of each function, and any pertinent non-obvious parts within them?
| return [] | ||
| # Extract information from relevant fields | ||
| data = [] | ||
| for k in e["role"].keys(): |
There was a problem hiding this comment.
Again, keys() is unnecessary, since you have an iterable already.
| scores = {} | ||
| keywords = {} | ||
|
|
||
| for k in self.vals.keys(): |
| if p["raw_skills"]: | ||
| return [p["raw_skills"]] |
There was a problem hiding this comment.
NB: For discussion. Here we make the assumption that each profile that contains 1 or more raw skills is 'complete', and that all other profiles are lacking skills. However, one might imagine that a profile with just one skill is incomplete and should have more skills generated, etc.
| for i in data: | ||
| for k in self.vals.keys(): | ||
| for v in self.vals[k]: |
There was a problem hiding this comment.
Consider whether there might be significant performance improvements through sets and set operations rather than checking every field individually. Maybe you should be breaking early if you find a keyword, since I don't know how helpful the 'score' is in practice. Can you take a look at the score variable and see how it does? I'd expect anything caught in 'roles' to get multiplied several times because many of the fields are copies or substrings of one another...
| class InferSkills: | ||
| def __init__(self): | ||
| self.vals = {} | ||
| with open('skillsets_labels.tsv', 'r') as f: |
There was a problem hiding this comment.
These are only the top-n skills related to a particular skill-set. We need to remember to expand this to the list of all skills with more than k occurrences.
| return ret | ||
|
|
||
|
|
||
| def load_sample_profile(self): |
There was a problem hiding this comment.
If you plan on using this in the final version, then you may want to rename it to something more general?
2 participants
No description provided.