import scipy
import numpy as np
import icgauge
from icgauge import experiment_frameworks
from icgauge import feature_extractors
run_experiment = True
if run_experiment:
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.train_and_dev,
assess_reader=icgauge.data_readers.test, #_official,
train_size=0.7,
phi_list=[icgauge.feature_extractors.simple_features],
class_func=icgauge.label_transformers.
identity_class_func, #ternary_class_func
train_func=icgauge.training_functions.
fit_logistic_at, #_with_crossvalidation,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=1)
# Print out the results
print "\n-- AFTER COMPLETION --"
print "Averaged correlation (95% CI): "
print np.round(np.mean(corr), 2), "+/-", np.round(np.std(corr), 2)
print "All correlations:"
print corr
print
print "Averaged Cronbach's alpha (95% CI): "
print np.round(np.mean(alpha), 2), "+/-", np.round(np.std(alpha), 2)
print " Labels: original 7-point scale"
print " Model: ordinal logistic model, all-threshold variant"
print " (ordinal classification, recommended based on"
print " experiments reported in Rennie and Srebro 2005"
print " and replicated with toy.json)"
print
corr, conf_matrix = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.toy,
assess_reader=None,
train_size=0.7,
phi_list=[
icgauge.feature_extractors.manual_content_flags,
icgauge.feature_extractors.length,
icgauge.feature_extractors.modal_presence,
icgauge.feature_extractors.hedge_presence,
icgauge.feature_extractors.conjunctives_presence,
icgauge.feature_extractors.punctuation_presence
],
class_func=icgauge.label_transformers.identity_class_func,
train_func=icgauge.training_functions.fit_logistic_at_with_crossvalidation,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=10)
print "Iterated result: "
print np.mean(corr)
print corr
print conf_matrix
print np.sum(conf_matrix)
"""
import scipy
import numpy as np
import icgauge
from icgauge import experiment_frameworks
from icgauge import feature_extractors
run_experiment = True
if run_experiment:
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.train_and_dev,
assess_reader=icgauge.data_readers.test,#_official,
train_size=0.7,
phi_list=[
icgauge.feature_extractors.simple_features
],
class_func=icgauge.label_transformers.identity_class_func, #ternary_class_func
train_func=icgauge.training_functions.fit_logistic_at,#_with_crossvalidation,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=1)
# Print out the results
print "\n-- AFTER COMPLETION --"
print "Averaged correlation (95% CI): "
print np.round(np.mean(corr),2), "+/-", np.round(np.std(corr),2)
print "All correlations:"
print corr
print
print "Averaged Cronbach's alpha (95% CI): "
print np.round(np.mean(alpha),2), "+/-", np.round(np.std(alpha),2)
import icgauge
from icgauge import experiment_frameworks
from icgauge import feature_extractors
run_experiment = True
if run_experiment:
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.train_and_dev,
assess_reader=None, #icgauge.data_readers.test_official,
train_size=0.7,
phi_list=[
# These two features were used to control for the effect of number
# of words and to get at the effect of the decomposition itself
# icgauge.feature_extractors.number_words_only,
# icgauge.feature_extractors.dimensional_decomposition,
# The real feature function for PCA, though, is:
icgauge.feature_extractors.semcom_pca_features
],
class_func=icgauge.label_transformers.identity_class_func, #ternary_class_func
train_func=icgauge.training_functions.fit_logistic_at,#_with_crossvalidation,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=1)
# Print out the results
print "\n-- AFTER COMPLETION --"
print "Averaged correlation (95% CI): "
print np.round(np.mean(corr),2), "+/-", np.round(np.std(corr),2)
print "All correlations:"
print corr
print " hedges, conjunctives, punctuation, determiner_usage"
print " Labels: original 7-point scale"
print " Model: ordinal logistic model, all-threshold variant"
print " (ordinal classification, recommended based on"
print " experiments reported in Rennie and Srebro 2005)"
print
corr, conf_matrix = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.toy,
assess_reader=None,
train_size=0.7,
phi_list=[icgauge.feature_extractors.manual_content_flags,
icgauge.feature_extractors.length,
icgauge.feature_extractors.modal_presence,
icgauge.feature_extractors.hedge_presence,
icgauge.feature_extractors.conjunctives_presence,
icgauge.feature_extractors.punctuation_presence,
icgauge.feature_extractors.determiner_usage],
class_func=icgauge.label_transformers.identity_class_func,
train_func=icgauge.training_functions.fit_logistic_at_with_crossvalidation,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=10)
print "Iterated result: "
print np.mean(corr)
print corr
print conf_matrix
print np.sum(conf_matrix)
"""
from icgauge import experiment_frameworks
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.train,
assess_reader=None,
train_size=0.7,
phi_list=[ # icgauge.feature_extractors.manual_content_flags,
# icgauge.feature_extractors.length,
# icgauge.feature_extractors.word_length_features,
# icgauge.feature_extractors.modal_presence,
# icgauge.feature_extractors.get_more_most_counts,
# icgauge.feature_extractors.hedge_presence,
# icgauge.feature_extractors.word_intensity,
# icgauge.feature_extractors.get_morphological_counts,
# icgauge.feature_extractors.transitional_presence,
# icgauge.feature_extractors.conjunctives_presence,
# icgauge.feature_extractors.punctuation_presence,
# icgauge.feature_extractors.determiner_usage,
# icgauge.feature_extractors.dimensional_decomposition,
# icgauge.feature_extractors.syntactic_parse_features,
# icgauge.feature_extractors.kannan_ambili,
icgauge.feature_extractors.semcom_sentiment_features
],
class_func=icgauge.label_transformers.
identity_class_func, #vs. ternary_class_func
train_func=icgauge.training_functions.
fit_logistic_at_with_crossvalidation, # does not have crossvalidation
score_func=scipy.stats.stats.pearsonr,
verbose=False)
# Print out the results
from icgauge import experiment_frameworks
corr, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.train,
assess_reader=None,
train_size=0.7,
phi_list=[icgauge.feature_extractors.manual_content_flags,
icgauge.feature_extractors.length,
icgauge.feature_extractors.word_length_features,
icgauge.feature_extractors.modal_presence,
icgauge.feature_extractors.get_more_most_counts,
icgauge.feature_extractors.hedge_presence,
icgauge.feature_extractors.word_intensity,
icgauge.feature_extractors.get_morphological_counts,
icgauge.feature_extractors.transitional_presence,
icgauge.feature_extractors.conjunctives_presence,
icgauge.feature_extractors.punctuation_presence,
icgauge.feature_extractors.determiner_usage,
icgauge.feature_extractors.dimensional_decomposition,
icgauge.feature_extractors.syntactic_parse_features,
icgauge.feature_extractors.kannan_ambili
],
class_func=icgauge.label_transformers.identity_class_func, #vs. ternary_class_func
train_func=icgauge.training_functions.fit_maxent_with_crossvalidation, # does not have crossvalidation
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=1)
# Print out the results
print "\n-- AFTER COMPLETION --"
print "Punctuation test framework:"
print " =================================== "
print " First test: dataset with punctuation"
print " =================================== "
print " Data: practice, test, toy dataset"
print " Features: all features"
print " Labels: original 7-point scale"
print " Model: logistic regression (classification)"
print
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.punctuated_set,
assess_reader=None,
train_size=0.7,
phi_list=[icgauge.feature_extractors.all_features],
class_func=icgauge.label_transformers.identity_class_func,
train_func=icgauge.training_functions.fit_logistic_at,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=10)
print "\n-- AFTER COMPLETION --"
print "Averaged correlation (95% CI): "
print np.round(np.mean(corr), 2), "+/-", np.round(np.std(corr), 2)
print "All correlations:"
print corr
print
print "Averaged Cronbach's alpha (95% CI): "
print np.round(np.mean(alpha), 2), "+/-", np.round(np.std(alpha), 2)
print "All alphas:"
print alpha
print "Punctuation test framework:"
print " =================================== "
print " First test: dataset with punctuation"
print " =================================== "
print " Data: practice, test, toy dataset"
print " Features: all features"
print " Labels: original 7-point scale"
print " Model: logistic regression (classification)"
print
corr, alpha, conf_matrix, details = experiment_frameworks.experiment_features_iterated(
train_reader=icgauge.data_readers.punctuated_set,
assess_reader=None,
train_size=0.7,
phi_list=[icgauge.feature_extractors.all_features],
class_func=icgauge.label_transformers.identity_class_func,
train_func=icgauge.training_functions.fit_logistic_at,
score_func=scipy.stats.stats.pearsonr,
verbose=False,
iterations=10)
print "\n-- AFTER COMPLETION --"
print "Averaged correlation (95% CI): "
print np.round(np.mean(corr),2), "+/-", np.round(np.std(corr),2)
print "All correlations:"
print corr
print
print "Averaged Cronbach's alpha (95% CI): "
print np.round(np.mean(alpha),2), "+/-", np.round(np.std(alpha),2)
print "All alphas:"