paranormal-or-skeptic/predict.py
2020-03-22 10:15:36 +01:00

49 lines
1.5 KiB
Python
Executable File

#!/usr/bin/python3
import pickle
import math
def clear_tokens(tokens):
tokens = tokens.replace('\n', ' ')
return tokens
def calc_post_prob(post, paranormal_class_logprob, sceptic_class_logprob, word_logprobs):
# dla kazdego tokenu z danego posta
text, timestap = post.rstrip('\n').split('\t')
text = clear_tokens(text)
tokens = text.lower().split(' ')
probs = {0.0 : 'sceptic', 0.0 : 'paranormal'}
for class_ in word_logprobs.keys():
product = 1
for token in tokens:
try:
product *= word_logprobs[class_][token]
except KeyError:
pass
# tu wzoru uzyj
if class_ == 'sceptic':
product *= sceptic_class_logprob
elif class_ == 'paranormal':
product *= paranormal_class_logprob
probs[abs(product)] = class_
print(probs)
return probs[max(probs.keys())]
def main():
with open('naive_base_model.pkl', 'rb') as f:
pickle_list = pickle.load(f)
paranormal_class_logprob = pickle_list[0]
sceptic_class_logprob = pickle_list[1]
word_logprobs = pickle_list[2]
with open('test-A/in.tsv') as in_f, open('test-A/out.tsv', 'w') as out_f:
for line in in_f:
hyp = calc_post_prob(line, paranormal_class_logprob, sceptic_class_logprob, word_logprobs)
if hyp == 'sceptic':
out_f.write(" S\n")
elif hyp == 'paranormal':
out_f.write(' P\n')
main()