andras/great-ai
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases 4 Packages Wiki Activity Actions

Fix explanations

Browse source
This commit is contained in:
Andras Schmelczer 2022-05-25 21:59:30 +02:00
parent 2d71fceb54
commit 06052732c0
2 changed files with 34 additions and 589 deletions
Download patch file Download diff file Expand all files Collapse all files

13
examples/predict_domain.py
View file

@ -1,5 +1,5 @@
import re import re
from typing import Dict, Iterable, List from typing import Dict, Iterable, List
from preprocess import preprocess from preprocess import preprocess
from pydantic import BaseModel from pydantic import BaseModel
@ -20,6 +20,8 @@ class DomainPrediction(BaseModel):
def predict_domain( def predict_domain(
text: str, model: Pipeline, cut_off_probability: float = 0.2 text: str, model: Pipeline, cut_off_probability: float = 0.2
) -> List[DomainPrediction]: ) -> List[DomainPrediction]:
assert 0 <= cut_off_probability <= 1
""" """
Predict the scientific domain of the input text. Predict the scientific domain of the input text.
Return labels until their sum likelihood is larger than cut_off_probability. Return labels until their sum likelihood is larger than cut_off_probability.
@ -42,10 +44,11 @@ def predict_domain(
results: List[DomainPrediction] = [] results: List[DomainPrediction] = []
for class_index, probability in best_classes: for class_index, probability in best_classes:
weights = model.named_steps["classifier"].feature_log_prob_[class_index] weights = model.named_steps["classifier"].feature_log_prob_[class_index]
domain = model.named_steps["classifier"].classes_[class_index]
results.append( results.append(
DomainPrediction( DomainPrediction(
domain=model.named_steps["classifier"].classes_[class_index], domain=domain,
probability=round(probability * 100), probability=round(probability * 100),
explanation=_get_explanation( explanation=_get_explanation(
feature_names=feature_names, feature_names=feature_names,
@ -69,11 +72,11 @@ def _get_explanation(
token_mapping: Dict[str, str], token_mapping: Dict[str, str],
) -> List[str]: ) -> List[str]:
influential = [ influential = [
(value * weight, name) (weight, name)
for name, value, weight in zip(feature_names, features, weights) for weight, value, name in zip(weights, features, feature_names)
if value if value
] ]
most_influential = sorted(influential, reverse=True)[:5] most_influential = sorted(influential, reverse=True)[:5]
return [token_mapping[v[1]] for v in most_influential] return [token_mapping[name] for _, name in most_influential]

610
examples/train.ipynb
View file

File diff suppressed because one or more lines are too long