def run_script(pdate_str, busjson, revjson, tipjson, senticsv, outfile):
# convert pdate to seconds since the epoch
pdate = du.date2int(du.str2date(pdate_str))
# load business objects
print 'Loading business objects from %s...' % busjson
all_buses, junk = ju.load_objects(busjson)
# load review objects
print 'loading review objects from %s...' % revjson
all_reviews, junk = ju.load_objects(revjson)
# load tip objects
print 'loading tip objects from %s...' % tipjson
all_tips, junk = ju.load_objects(tipjson)
# load sentiment ranking data derived from tip and review data
print 'loading sentiment rankings from %s...' % senticsv
all_senti = cu.load_matrix(senticsv, has_hdr=False)
# generate a data set the specified prediction date
print('generate data set for prediction date %s...' % pdate_str)
buses = du.gen_dataset(pdate, all_buses, all_reviews, all_tips, all_senti)
# write data set to file
print('writing generated data set to %s...' % outfile)
ju.save_objects(buses, outfile)
def run_script(busjson, revjson, tipjson, senticsv, init_pdate, delta, ctype=linsvm,
usamp=True, binary=None, rfe=False, pca=-1, reg=False, feat_info=fi.data_feat_info,
states=None):
print 'Initial prediction date: %s' % init_pdate
print 'Time delta: %d months' % delta
if (states):
print 'limiting data to restaurants in: %s' % str(states)
# convert pdate to secondds since the epoch
pdate = du.date2int(du.str2date(init_pdate))
# load business objects
print 'Loading business objects from %s...' % busjson
all_buses, junk = ju.load_objects(busjson)
# load review objects
print 'loading review objects from %s...' % revjson
all_reviews, junk = ju.load_objects(revjson)
# load tip objects
print 'loading tip objects from %s...' % tipjson
all_tips, junk = ju.load_objects(tipjson)
# load sentiment ranking data derived from tip and review data
print 'loading sentiment rankings from %s...' % senticsv
all_senti = cu.load_matrix(senticsv, has_hdr=False)
# reduce the number of features using recursive feature elimination
# - See http://scikit-learn.org/stable/auto_examples/plot_rfe_with_cross_validation.html#example-plot-rfe-with-cross-validation-py
# - See http://stackoverflow.com/questions/23815938/recursive-feature-elimination-and-grid-search-using-scikit-learn
if (reg):
# create the least squares linear regressor
print 'using least squares linear regression...'
c = linmod.LinearRegression()
# grid search not supported for linear regression (???)
param_grid = None
elif (ctype==rbfsvm):
# create RBF SVM to test
#c = svm.NuSVC(kernel='rbf')
c = svm.SVC(kernel='rbf')
# configure parameter grid for grid search
C_range = 10.0 ** np.arange(-3, 5)
gamma_range = 10.0 ** np.arange(-4, 3)
if (rfe):
print 'RFE not currently supported for RBF SVM...'
#c = fs.RFECV(c, step=1)
#pgrid = []
#for C in C_range:
# for gamma in gamma_range:
# pgrid.append({'C':C,'gamma':gamma})
#pgrid = [{'gamma':0.5},{'gamma':0.1},{'gamma':0.01},{'gamma':0.001},{'gamma':0.0001}]
#param_grid = {'estimator_params': pgrid}
print 'using RBF SVM...'
param_grid = dict(gamma=gamma_range, C=C_range)
elif (ctype==knn):
# create a KNN classifier
c = neigh.KNeighborsClassifier()
if (rfe):
print 'RFE not currently supported for k-nearesrt neighbors...'
print 'using k-mearest neighbors...'
param_grid = {'n_neighbors':[1,2,3,4,5,6,7,8,9,10,15,20,25,30],
'weights':['uniform','distance'],
'p':[1,2,3,4,5,6,7,8,9,10]}
elif (ctype==ada):
# create boosted classifier
c = ensemble.AdaBoostClassifier()
if (rfe):
print 'RFE not currently supported for AdaBoost...'
print 'using AdaBoost...'
param_grid = {'n_estimators':[5, 10, 25, 40, 50, 60, 75, 85, 100],
'learning_rate':[0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0]}
elif (ctype==rf):
# create random forest classifier
c = ensemble.RandomForestClassifier()
if (rfe):
print 'RFE not currently supported for random forest...'
print 'using random forest...'
param_grid = {'n_estimators':[5, 10, 25, 40, 50, 60, 75, 85, 100],
'max_depth':[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, None]}
elif (ctype==dt):
# create decision tree classifier
c = tree.DecisionTreeClassifier()
# max feats - subtract 1 because data feats includes the class label
if (rfe):
print 'RFE not supported with decision trees...'
print 'using decision tree...'
param_grid = {'max_depth': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, None]}
else:
# create linear SVM to test
c = svm.LinearSVC()
# configure parameter grid for grid search
C_range = 10.0 ** np.arange(-3, 5)
if (rfe):
print 'using linear SVM with RFE...'
c = fs.RFECV(c, step=1)
pgrid = []
for C in C_range:
pgrid.append({'C':C})
#pgrid = [{'C':0.01},{'C':0.1},{'C':1},{'C':10},{'C':100},{'C':1000},{'C':10000}]
param_grid = {'estimator_params': pgrid}
else:
print 'using linear SVM...'
param_grid = {'C': C_range}
# run the walk-forward cross validation and collect the results
print('run walk-forward cross validation...')
if (usamp):
print(' under-sampling still open class...')
else:
print(' NOT under-sampling still open class...')
results = wfcvutils.wfcv(c, param_grid, all_buses, all_reviews, all_tips, all_senti,
pdate, delta*du.month, pca=pca, usamp=usamp,
binary=binary, reg=reg, feat_info=feat_info, states=states)
# combine the results to produce overall metrics
y_true = None
y_pred = None
for r in results:
if (y_true is None):
y_true = r[0]
else:
y_true = np.hstack((y_true, r[0]))
if (y_pred is None):
y_pred = r[1]
else:
y_pred = np.hstack((y_pred, r[1]))
# print out an overall classification report
print('\n=========================================')
print('Overall metrics for all prediction dates:\n')
if (len(results) != 0):
if (reg):
wfcvutils.print_reg_metrics(y_true, y_pred)
else:
cm = metrics.confusion_matrix(y_true, y_pred)
wfcvutils.print_cm(cm)
#print(metrics.classification_report(y_true, y_pred, target_names=fi.class_names))
else:
print ' NO RESULTS\n'
