# -*- coding: utf-8 -*-
"""
Created on Mon Nov 16 18:35:09 2015
@author: Valou
"""
import os
from extraction import extract2CRFsuite
path = "/Users/Valou/Documents/TELECOM_PARISTECH/Stage_Lucas/Datasets/Semaine/"
ALL_LABELS = {'attitude_positive', 'attitude_negative', 'source', 'target'}
ALL_FILES = sorted(os.listdir(path+"all/dump/")) # nom de tous les fichiers contenus dans path+"all/dump" tries dans l'ordre
label = 'attitude'
label_select = 'attitude'
for i in range(1): # i represente une session ?
filename = ALL_FILES[i]
X, y = extract2CRFsuite(path+"all/dump/"+filename,
path+"all/dump_audio/"+filename,
path+"all/dump_mfcc/"+filename,
label, 'TEXT')
def cvloo(label, path_results, params, label_select=None, LOOP_TEST=False, valence=False):
u"""Compute the Cross-validation for the given label.
valence is True if we wanna distinguish the positive and negative attitudes
"""
if label_select is None:
label_select = label
opt = params["opt"]
truepos_o, falsepos_o, falseneg_o = (0, 0, 0)
precision = {}
recall = {}
trainer = pycrfsuite.Trainer(verbose=False)
for i in range(len(ALL_FILES)):
filename = ALL_FILES[i]
X, y = extract2CRFsuite(
path + "all/dump" + valence * "_attitudeposneg_only" + "/" + filename,
path + "all/dump_audio/" + filename,
path + "all/dump_mfcc/" + filename,
label,
params,
)
for x_seq, y_seq in zip(X, y):
trainer.append(x_seq, y_seq, i)
trainer.set_params(
{
"c1": params["c1"], # coefficient for L1 penalty
"c2": params["c2"], # coefficient for L2 penalty
"max_iterations": params["max_it"], # stop earlier
# include transitions that are possible, but not observed
"feature.possible_transitions": False,
}
)
# print("Beginning of the training")
for i in range(len(ALL_FILES)):
# for i in range(1):
filename = ALL_FILES[i]
filename_model = filename.split(".")[0] # to threw away the extension
# Training
trainer.train(path_model + "model_%s_" % opt + filename_model, i)
# Testing
X_test, y_test = extract2CRFsuite(
path + "all/dump" + valence * "_attitudeposneg_only" + "/" + filename,
path + "all/dump_audio/" + filename,
path + "all/dump_mfcc/" + filename,
label,
params,
)
tagger = pycrfsuite.Tagger(verbose=False)
tagger.open(path_model + "model_%s_" % opt + filename_model)
truepos, falsepos, falseneg = (0, 0, 0)
for sent, corr_labels in zip(X_test, y_test):
pred_labels = tagger.tag(sent)
trueposAdd, falseposAdd, falsenegAdd = F1_token(pred_labels, corr_labels, label_select)
truepos += trueposAdd
falsepos += falseposAdd
falseneg += falsenegAdd
precision[filename] = "%.2f" % (truepos / (truepos + falsepos + 0.01) * 100)
recall[filename] = "%.2f" % (truepos / (truepos + falseneg + 0.01) * 100)
truepos_o += truepos
falsepos_o += falsepos
falseneg_o += falseneg
precision["overall"] = "%.2f" % (truepos_o / (truepos_o + falsepos_o + 0.01) * 100)
recall["overall"] = "%.2f" % (truepos_o / (truepos_o + falseneg_o + 0.01) * 100)
F1 = (
2
* float(precision["overall"])
* float(recall["overall"])
/ (float(precision["overall"]) + float(recall["overall"]) + 1e-5)
)
# If there is pos and neg differentiation for the attitudes
if valence == True and label.__class__ == list:
label = "attitud_posneg"
# Dump the different results on results
ext = ".txt"
dump_resultats(precision, recall, F1, path_results + "results_CVLOO_%s_" % (opt) + label + "_" + label_select + ext)
if LOOP_TEST: # if loop test dump the ALL the results in 1 file
dump_resultats_total(
precision, recall, F1, path_results + "results_total_%s_" % (opt) + label + "_" + label_select + ext, params
)
return_sent = "Precision : %s, Recall : %s, F1 : %.2f" % (precision["overall"], recall["overall"], F1)
return return_sent
