def genre_gridsearch(modelname, c_range, ftstart, ftend, ftstep, positive_tags = ['elite'], negative_tags = ['vulgar'], excl_below = 1700, excl_above = 2000):
# Function does a gridsearch to identify an optimal number of features and setting of
# the regularization constant; then produces that model.
sourcefolder = '/Users/tunder/Dropbox/GenreProject/python/reception/fiction/fromEF/'
extension = '.tsv'
#metadatapath = '/Users/tunder/Dropbox/GenreProject/python/reception/fiction/prestigeficmeta.csv'
metadatapath = '/Users/tunder/Dropbox/GenreProject/python/reception/fiction/littlemagazines.csv'
vocabpath = '/Users/tunder/Dropbox/fiction/lexicon/' + modelname + '.txt'
if os.path.exists(vocabpath):
print('Vocabulary for ' + modelname + ' already exists. Using it.')
outputpath = '/Users/tunder/Dropbox/GenreProject/python/reception/fiction/' + modelname + '.csv'
# We can simply exclude volumes from consideration on the basis on any
# metadata category we want, using the dictionaries defined below.
## EXCLUSIONS.
excludeif = dict()
excludeifnot = dict()
excludeabove = dict()
excludebelow = dict()
excludebelow['firstpub'] = excl_below
excludeabove['firstpub'] = excl_above
sizecap = 700
# CLASSIFY CONDITIONS
# print()
# print("You can also specify positive tags to be excluded from training, and/or a pair")
# print("of integer dates outside of which vols should be excluded from training.")
# print("If you add 'donotmatch' to the list of tags, these volumes will not be")
# print("matched with corresponding negative volumes.")
# print()
# ## testphrase = input("Comma-separated list of such tags: ")
testphrase = ''
testconditions = set([x.strip() for x in testphrase.split(',') if len(x) > 0])
datetype = "firstpub"
numfeatures = ftend
regularization = .000075
# linting the code would get rid of regularization, which is at this
# point an unused dummy parameter
paths = (sourcefolder, extension, metadatapath, outputpath, vocabpath)
exclusions = (excludeif, excludeifnot, excludebelow, excludeabove, sizecap)
classifyconditions = (positive_tags, negative_tags, datetype, numfeatures, regularization, testconditions)
modelparams = 'logistic', 12, ftstart, ftend, ftstep, c_range
matrix, rawaccuracy, allvolumes, coefficientuples = train.tune_a_model(paths, exclusions, classifyconditions, modelparams)
print('If we divide the dataset with a horizontal line at 0.5, accuracy is: ', str(rawaccuracy))
tiltaccuracy = train.diachronic_tilt(allvolumes, 'linear', [])
print("Divided with a line fit to the data trend, it's ", str(tiltaccuracy))
def genre_gridsearch(metadatapath, modelname, c_range, ftstart, ftend, ftstep, positive_tags = ['fic'], negative_tags = ['bio'], excl_below = 1700, excl_above = 2000):
# Function does a gridsearch to identify an optimal number of features and setting of
# the regularization constant; then produces that model.
sourcefolder = '../sourcefiles/'
extension = '.tsv'
vocabpath = '../lexicons/' + modelname + '.txt'
if os.path.exists(vocabpath):
print('Vocabulary for ' + modelname + ' already exists. Using it.')
outputpath = '../modeloutput/' + modelname + '.csv'
# We can simply exclude volumes from consideration on the basis on any
# metadata category we want, using the dictionaries defined below.
## EXCLUSIONS.
excludeif = dict()
excludeifnot = dict()
excludeabove = dict()
excludebelow = dict()
excludebelow['firstpub'] = excl_below
excludeabove['firstpub'] = excl_above
sizecap = 75
# CLASSIFY CONDITIONS
# print()
# print("You can also specify positive tags to be excluded from training, and/or a pair")
# print("of integer dates outside of which vols should be excluded from training.")
# print("If you add 'donotmatch' to the list of tags, these volumes will not be")
# print("matched with corresponding negative volumes.")
# print()
# ## testphrase = input("Comma-separated list of such tags: ")
testphrase = ''
testconditions = set([x.strip() for x in testphrase.split(',') if len(x) > 0])
datetype = "firstpub"
numfeatures = ftend
regularization = .000075
# linting the code would get rid of regularization, which is at this
# point an unused dummy parameter
paths = (sourcefolder, extension, metadatapath, outputpath, vocabpath)
exclusions = (excludeif, excludeifnot, excludebelow, excludeabove, sizecap)
classifyconditions = (positive_tags, negative_tags, datetype, numfeatures, regularization, testconditions)
modelparams = 'logistic', 12, ftstart, ftend, ftstep, c_range
matrix, rawaccuracy, allvolumes, coefficientuples = train.tune_a_model(paths, exclusions, classifyconditions, modelparams)
print('If we divide the dataset with a horizontal line at 0.5, accuracy is: ', str(rawaccuracy))
tiltaccuracy = train.diachronic_tilt(allvolumes, 'linear', [])
print("Divided with a line fit to the data trend, it's ", str(tiltaccuracy))
def gridsearch_a_model(metadatapath,
sourcefolder,
c_range,
ftstart,
ftend,
ftstep,
positive_tags=['f'],
negative_tags=['m']):
''' Function does a gridsearch to identify an optimal number of features and setting of
the regularization constant; then produces that model. Note that we do not use this for
models of specific decades. Just initially for model selection.'''
modelname = metadatapath.replace('.//models/', '').replace('_meta.csv', '')
extension = '.tsv'
vocabpath = metadatapath.replace('_meta', '_vocab')
if os.path.exists(vocabpath):
print('Vocabulary for ' + modelname + ' already exists. Using it.')
outputpath = metadatapath.replace('_meta', '')
## EXCLUSIONS. # not used in this project
excludeif = dict()
excludeifnot = dict()
excludeabove = dict()
excludebelow = dict()
sizecap = 2000
# CLASSIFY CONDITIONS # not used in this project
testconditions = set()
datetype = "firstpub"
numfeatures = ftend
regularization = .000075
# linting the code would get rid of regularization, which is at this
# point an unused dummy parameter
paths = (sourcefolder, extension, metadatapath, outputpath, vocabpath)
exclusions = (excludeif, excludeifnot, excludebelow, excludeabove, sizecap)
classifyconditions = (positive_tags, negative_tags, datetype, numfeatures,
regularization, testconditions)
modelparams = 'logistic', 12, ftstart, ftend, ftstep, c_range
matrix, rawaccuracy, allvolumes, coefficientuples = train.tune_a_model(
paths, exclusions, classifyconditions, modelparams)
print(
'If we divide the dataset with a horizontal line at 0.5, accuracy is: ',
str(rawaccuracy))
tiltaccuracy = train.diachronic_tilt(allvolumes, 'linear', [])
print("Divided with a line fit to the data trend, it's ",
str(tiltaccuracy))
def genre_gridsearch(modelname, c_range, ftstart, ftend, ftstep, positive_tags, negative_tags = ['random', 'grandom', 'chirandom'], excl_below = 1700, excl_above = 2000, metadatapath = '../metadata/concatenatedmeta.csv'):
'''
This is at present the workhorse function of the script. It does a
gridsearch to identify an optimal number of features and setting of
the regularization constant. Then it produces that model, and writes it
to disk.
'''
#sourcefolder = '/Users/tunder/Dropbox/fiction/newtsvs'
sourcefolder = '../sourcefiles'
extension = '.tsv'
vocabpath = '../lexicons/' + modelname + '.txt'
if os.path.exists(vocabpath):
print('Vocabulary for ' + modelname + ' already exists. Using it.')
outputpath = '../modeloutput/' + modelname + '.csv'
# We can simply exclude volumes from consideration on the basis on any
# metadata category we want, using the dictionaries defined below.
## EXCLUSIONS.
excludeif = dict()
excludeifnot = dict()
excludeabove = dict()
excludebelow = dict()
excludebelow['firstpub'] = excl_below
excludeabove['firstpub'] = excl_above
sizecap = 400
# CLASSIFY CONDITIONS
testphrase = ''
testconditions = set([x.strip() for x in testphrase.split(',') if len(x) > 0])
datetype = "firstpub"
numfeatures = ftend
regularization = .000075
# linting the code would get rid of regularization, which is at this
# point an unused dummy parameter
paths = (sourcefolder, extension, metadatapath, outputpath, vocabpath)
exclusions = (excludeif, excludeifnot, excludebelow, excludeabove, sizecap)
classifyconditions = (positive_tags, negative_tags, datetype, numfeatures, regularization, testconditions)
modelparams = 'logistic', 10, ftstart, ftend, ftstep, c_range
matrix, rawaccuracy, allvolumes, coefficientuples = train.tune_a_model(paths, exclusions, classifyconditions, modelparams)
print('If we divide the dataset with a horizontal line at 0.5, accuracy is: ', str(rawaccuracy))
