def run_unittests(X_test, y_test, verbose=True):
print 'Do some unit tests on seriesmodel, featurizer...'
print '0) Set-up (prepare data)'
sm = SeriesModel(reference_time=9)
if True:
print 'A) seriesmodel preprocessing'
if verbose:
print X_test.iloc[0][0:5, 0:4]
print 'DI, reftime = 2'
DI = sm.preprocess(X_test)
if verbose:
print DI.iloc[0][0:5, 0:4]
print 'DII, reftime = 3'
sm = SeriesModel(color_vector_type='DII', reference_time=3)
DII = sm.preprocess(X_test)
if verbose:
print DII.iloc[0][0:5, 0:4]
sm.confusion_labels = sm._build_confusion_labels(y_test)
Z = sm._prepare_data(X_test)
if True:
print 'B) Set-up results dataframes'
if verbose:
print 'Predictions'
print sm.predictions.head()
print 'Probabilities'
print sm.probabilities.head()
print 'Metrics'
print sm.metrics
print 'Confusion Labels'
print sm.confusion_labels
print 'Scores'
print sm.scores
print '1) Fit one timestep...'
sm._fit_one_timestep(Z, y_test, 10)
print 'Fit another'
sm._fit_one_timestep(Z, y_test, 20)
print 'And another'
sm._fit_one_timestep(Z, y_test, 30)
print 'Predict these timesteps'
sm.times = [10,20,30]
yd,yg,yc = sm.predict(X_test)
print 'Score timesteps'
results = sm.score(y_test, verbose=True)
print 'Try again using the full workthrough'
sm = SeriesModel(reference_time=9, min_time=3)
sm.fit(X_test, y_test, verbose=True, debug=True)
yd, yg, yc = sm.predict(X_test.iloc[0:10])
results = sm.score(y_test.iloc[0:10])
if verbose:
print 'Check shape of results'
for c in sm.results.columns:
print c, sm.results[c].iloc[0].shape
print 'Check shape of predictions'
for k, v in sm.predictions.iteritems():
print k, len(v)
print 'Check shape of probabilities'
for k, v in sm.probabilities.iteritems():
print k, v.shape
print 'Check confusion matrices for sense'
print sm.confusion_labels['gram']
print sm.scores['gram']['micro'].iloc[0]['confusion_matrix']
print sm.scores['gram']['n'].iloc[0]['confusion_matrix']
if True:
print 'A) Subset data'
nt = 4
X_sub = sm._subset_data(Z, nt)
if verbose:
print X_sub.shape, X_sub.iloc[0].shape, Z.shape, Z.iloc[0].shape
print Z.iloc[0][0:5,0:5]
print X_sub.iloc[0][:,0:5]
print 'B) Featurize data'
Xf, featurizer = sm._featurize_class(X_sub, 'poly', {'n':3})
print 'i) Try all the same featurizer'
Xd, Xg, Xc = sm._featurize(X_sub, nt)
if verbose:
print Xd.head()
print Xg.head()
print Xc.head()
print 'ii) Try all different featurizers'
sm.gram_base_featurizer_arguments = {'n':6}
sm.gram_base_featurizer = 'poly'
sm.classification_base_featurizer = None
Xd2, Xg2, Xc2 = sm._featurize(X_sub, nt)
if verbose:
print Xd2.head()
print Xg2.head()
print Xc2.head()
print 'C) Run detection model'
nt = 30
X_sub = sm._subset_data(Z, nt)
Xd, Xg, Xc = sm._featurize(X_sub, nt)
print 'i) Convert featurized data to np array'
np_Xd = sm._pandas_to_numpy(Xd)
if verbose:
print np_Xd.shape
print y_test['detection'].values.shape
print 'ii) Train model'
detection_model = sm._fit_class(np_Xd, y_test['detection'].values,
'LR',
sm.detection_base_model_arguments,
step=('detection t=%d' % nt))
if verbose:
print 'Inspect stored models, featurizers'
print sm.models
print sm.featurizers
if True:
print '\n\n2) PolynomialFeaturizer'
start = time.time()
PF = PolynomialFeaturizer(n=4, reference_time=2, verbose=True)
mycoefs, myscores = PF.fit_transform(DI)
# myscores = PF.score()
DI_pred = PF.predict(DI, mycoefs)
end = time.time()
print 'Featurized, predicted %d test trails in %d seconds' % (len(y_test),(end-start))
# need to add curve visualization here to do a sanity check
# on the data
print 'Curve visualization...'
STP = SpotTimePlot(y_test, used_column_headers)
STP.plot_fits(DI, DI_pred)
return sm
def run_unittests(X_test, y_test, verbose=True):
print 'Do some unit tests on seriesmodel, featurizer...'
print '0) Set-up (prepare data)'
sm = SeriesModel(reference_time=9)
if True:
print 'A) seriesmodel preprocessing'
if verbose:
print X_test.iloc[0][0:5, 0:4]
print 'DI, reftime = 2'
DI = sm.preprocess(X_test)
if verbose:
print DI.iloc[0][0:5, 0:4]
print 'DII, reftime = 3'
sm = SeriesModel(color_vector_type='DII', reference_time=3)
DII = sm.preprocess(X_test)
if verbose:
print DII.iloc[0][0:5, 0:4]
sm.confusion_labels = sm._build_confusion_labels(y_test)
Z = sm._prepare_data(X_test)
if True:
print 'B) Set-up results dataframes'
if verbose:
print 'Predictions'
print sm.predictions.head()
print 'Probabilities'
print sm.probabilities.head()
print 'Metrics'
print sm.metrics
print 'Confusion Labels'
print sm.confusion_labels
print 'Scores'
print sm.scores
print '1) Fit one timestep...'
sm._fit_one_timestep(Z, y_test, 10)
print 'Fit another'
sm._fit_one_timestep(Z, y_test, 20)
print 'And another'
sm._fit_one_timestep(Z, y_test, 30)
print 'Predict these timesteps'
sm.times = [10, 20, 30]
yd, yg, yc = sm.predict(X_test)
print 'Score timesteps'
results = sm.score(y_test, verbose=True)
print 'Try again using the full workthrough'
sm = SeriesModel(reference_time=9, min_time=3)
sm.fit(X_test, y_test, verbose=True, debug=True)
yd, yg, yc = sm.predict(X_test.iloc[0:10])
results = sm.score(y_test.iloc[0:10])
if verbose:
print 'Check shape of results'
for c in sm.results.columns:
print c, sm.results[c].iloc[0].shape
print 'Check shape of predictions'
for k, v in sm.predictions.iteritems():
print k, len(v)
print 'Check shape of probabilities'
for k, v in sm.probabilities.iteritems():
print k, v.shape
print 'Check confusion matrices for sense'
print sm.confusion_labels['gram']
print sm.scores['gram']['micro'].iloc[0]['confusion_matrix']
print sm.scores['gram']['n'].iloc[0]['confusion_matrix']
if True:
print 'A) Subset data'
nt = 4
X_sub = sm._subset_data(Z, nt)
if verbose:
print X_sub.shape, X_sub.iloc[0].shape, Z.shape, Z.iloc[0].shape
print Z.iloc[0][0:5, 0:5]
print X_sub.iloc[0][:, 0:5]
print 'B) Featurize data'
Xf, featurizer = sm._featurize_class(X_sub, 'poly', {'n': 3})
print 'i) Try all the same featurizer'
Xd, Xg, Xc = sm._featurize(X_sub, nt)
if verbose:
print Xd.head()
print Xg.head()
print Xc.head()
print 'ii) Try all different featurizers'
sm.gram_base_featurizer_arguments = {'n': 6}
sm.gram_base_featurizer = 'poly'
sm.classification_base_featurizer = None
Xd2, Xg2, Xc2 = sm._featurize(X_sub, nt)
if verbose:
print Xd2.head()
print Xg2.head()
print Xc2.head()
print 'C) Run detection model'
nt = 30
X_sub = sm._subset_data(Z, nt)
Xd, Xg, Xc = sm._featurize(X_sub, nt)
print 'i) Convert featurized data to np array'
np_Xd = sm._pandas_to_numpy(Xd)
if verbose:
print np_Xd.shape
print y_test['detection'].values.shape
print 'ii) Train model'
detection_model = sm._fit_class(np_Xd,
y_test['detection'].values,
'LR',
sm.detection_base_model_arguments,
step=('detection t=%d' % nt))
if verbose:
print 'Inspect stored models, featurizers'
print sm.models
print sm.featurizers
if True:
print '\n\n2) PolynomialFeaturizer'
start = time.time()
PF = PolynomialFeaturizer(n=4, reference_time=2, verbose=True)
mycoefs, myscores = PF.fit_transform(DI)
# myscores = PF.score()
DI_pred = PF.predict(DI, mycoefs)
end = time.time()
print 'Featurized, predicted %d test trails in %d seconds' % (
len(y_test), (end - start))
# need to add curve visualization here to do a sanity check
# on the data
print 'Curve visualization...'
STP = SpotTimePlot(y_test, used_column_headers)
STP.plot_fits(DI, DI_pred)
return sm
from featurizer import PolynomialFeaturizer
from timeseriesplotter import SpotTimePlot
myfile = open('pickles/DI.pkl', 'rb')
X = pickle.load(myfile)
myfile.close()
# print len(X), X.iloc[0].shape
myfile = open('pickles/used_column_headers.pkl', 'rb')
column_headers = pickle.load(myfile)
myfile.close()
# print len(column_headers)
myfile = open('pickles/ydf.pkl', 'rb')
y = pickle.load(myfile)
myfile.close()
# print y.head()
# sm = SeriesModel(reference_time=9, min_time=3)
# PF = PolynomialFeaturizer(n=4, reference_time=9, verbose=True)
# mycoefs, myscores = PF.fit_transform(X)
#
#
# DI_pred = PF.predict(X, mycoefs)
STP = SpotTimePlot(y, column_headers)
# STP.plot_raws(X, averages=False)
STP.plot_raws(X)