def test_rus_fit_transform():
"""Test the fit transform routine"""
# Resample the data
rus = RandomUnderSampler(random_state=RND_SEED)
X_resampled, y_resampled = rus.fit_transform(X, Y)
currdir = os.path.dirname(os.path.abspath(__file__))
X_gt = np.load(os.path.join(currdir, 'data', 'rus_x.npy'))
y_gt = np.load(os.path.join(currdir, 'data', 'rus_y.npy'))
assert_array_equal(X_resampled, X_gt)
assert_array_equal(y_resampled, y_gt)
def test_rus_fit_transform():
"""Test the fit transform routine"""
# Resample the data
rus = RandomUnderSampler(random_state=RND_SEED)
X_resampled, y_resampled = rus.fit_transform(X, Y)
currdir = os.path.dirname(os.path.abspath(__file__))
X_gt = np.load(os.path.join(currdir, 'data', 'rus_x.npy'))
y_gt = np.load(os.path.join(currdir, 'data', 'rus_y.npy'))
assert_array_equal(X_resampled, X_gt)
assert_array_equal(y_resampled, y_gt)
def test_rus_fit():
"""Test the fitting method"""
# Create the object
rus = RandomUnderSampler(random_state=RND_SEED)
# Fit the data
rus.fit(X, Y)
# Check if the data information have been computed
assert_equal(rus.min_c_, 0)
assert_equal(rus.maj_c_, 1)
assert_equal(rus.stats_c_[0], 500)
assert_equal(rus.stats_c_[1], 4500)
def test_rus_fit():
"""Test the fitting method"""
# Create the object
rus = RandomUnderSampler(random_state=RND_SEED)
# Fit the data
rus.fit(X, Y)
# Check if the data information have been computed
assert_equal(rus.min_c_, 0)
assert_equal(rus.maj_c_, 1)
assert_equal(rus.stats_c_[0], 500)
assert_equal(rus.stats_c_[1], 4500)
def test_rus_fit_transform_with_indices():
"""Test the fit transform routine with indices support"""
# Resample the data
rus = RandomUnderSampler(return_indices=True, random_state=RND_SEED)
X_resampled, y_resampled, idx_under = rus.fit_transform(X, Y)
currdir = os.path.dirname(os.path.abspath(__file__))
X_gt = np.load(os.path.join(currdir, 'data', 'rus_x.npy'))
y_gt = np.load(os.path.join(currdir, 'data', 'rus_y.npy'))
idx_gt = np.load(os.path.join(currdir, 'data', 'rus_idx.npy'))
assert_array_equal(X_resampled, X_gt)
assert_array_equal(y_resampled, y_gt)
assert_array_equal(idx_under, idx_gt)
def test_rus_fit_transform_with_indices():
"""Test the fit transform routine with indices support"""
# Resample the data
rus = RandomUnderSampler(return_indices=True, random_state=RND_SEED)
X_resampled, y_resampled, idx_under = rus.fit_transform(X, Y)
currdir = os.path.dirname(os.path.abspath(__file__))
X_gt = np.load(os.path.join(currdir, 'data', 'rus_x.npy'))
y_gt = np.load(os.path.join(currdir, 'data', 'rus_y.npy'))
idx_gt = np.load(os.path.join(currdir, 'data', 'rus_idx.npy'))
assert_array_equal(X_resampled, X_gt)
assert_array_equal(y_resampled, y_gt)
assert_array_equal(idx_under, idx_gt)
def test_rus_transform_wt_fit():
"""Test either if an error is raised when transform is called before
fitting"""
# Create the object
rus = RandomUnderSampler(random_state=RND_SEED)
assert_raises(RuntimeError, rus.transform, X, Y)
def train(self, data):
self._cls = svm.SVC(C=self._complexity,
kernel='linear',
probability=self._prob_enabled)
# Data unchanged if resample type 'none' or 'cls-wgt'
X_resample = data.X
y_resample = data.y
if self._resample_type == 'cls-wgt':
self._cls.class_weight = 'balanced'
elif self._resample_type == 'over':
resample = RandomOverSampler(ratio='auto',
random_state=self._resample_seed)
resample.verbose = False # Req. due to mistake in library
X_resample, y_resample = resample.fit_transform(data.X, data.y)
elif self._resample_type == 'under':
resample = RandomUnderSampler(ratio='auto',
random_state=self._resample_seed,
verbose=False)
X_resample, y_resample = resample.fit_transform(data.X, data.y)
if self._norm_type == 'std':
self._scaler = preprocessing.StandardScaler().fit(X_resample)
else:
self._scaler = preprocessing.MinMaxScaler().fit(X_resample)
X_norm = self._scaler.transform(X_resample)
self._cls.fit(X_norm, y_resample)
def test_rus_fit_invalid_ratio():
"""Test either if an error is raised when the balancing ratio to fit is
smaller than the one of the data"""
# Create the object
ratio = 1. / 10000.
rus = RandomUnderSampler(ratio=ratio, random_state=RND_SEED)
# Fit the data
assert_raises(RuntimeError, rus.fit, X, Y)
def test_rus_fit_single_class():
"""Test either if an error when there is a single class"""
# Create the object
rus = RandomUnderSampler(random_state=RND_SEED)
# Resample the data
# Create a wrong y
y_single_class = np.zeros((X.shape[0], ))
assert_raises(RuntimeError, rus.fit, X, y_single_class)
def test_rus_init():
"""Test the initialisation of the object"""
# Define a ratio
verbose = True
ratio = 'auto'
rus = RandomUnderSampler(ratio=ratio,
random_state=RND_SEED,
verbose=verbose)
assert_equal(rus.rs_, RND_SEED)
assert_equal(rus.verbose, verbose)
assert_equal(rus.min_c_, None)
assert_equal(rus.maj_c_, None)
assert_equal(rus.stats_c_, {})
weights=[0.1, 0.9],
n_informative=3,
n_redundant=1,
flip_y=0,
n_features=20,
n_clusters_per_class=1,
n_samples=5000,
random_state=10)
# Instanciate a PCA object for the sake of easy visualisation
pca = PCA(n_components=2)
# Fit and transform x to visualise inside a 2D feature space
X_vis = pca.fit_transform(X)
# Apply the random under-sampling
rus = RandomUnderSampler()
X_resampled, y_resampled = rus.fit_transform(X, y)
X_res_vis = pca.transform(X_resampled)
# Two subplots, unpack the axes array immediately
f, (ax1, ax2) = plt.subplots(1, 2)
ax1.scatter(X_vis[y == 0, 0],
X_vis[y == 0, 1],
label="Class #0",
alpha=0.5,
edgecolor=almost_black,
facecolor=palette[0],
linewidth=0.15)
ax1.scatter(X_vis[y == 1, 0],
X_vis[y == 1, 1],
def preProcess(theFileName):
df = pd.read_csv(str(theFileName))
if 'Unnamed: 0' in df.columns:
df = df.drop('Unnamed: 0', axis=1)
labBin = sklearn.preprocessing.LabelBinarizer()
df['y'] = labBin.fit_transform(df['y'])
dp = pd.get_dummies(df)
X = dp.drop('y', axis=1)
y = dp[['y']]
# get the features
theFeatures = X.columns
# convert the dataframes to arrays
X = X.values
y = y.values
y.shape = np.shape(y)[0]
yOrig = y[:] # need this later for plotting feature impacts
# and carry out feature scaling
X = StandardScaler().fit_transform(X)
#=======================================================================
# apply random undersampling if labels are imbalanced
labelSkewness = 100 * np.sum(y) * 1. / np.shape(y)[0]
if np.min([labelSkewness, 100 - labelSkewness]) < (100. / 3.):
rus = RandomUnderSampler(verbose=0)
X, y = rus.fit_sample(X, y)
#=======================================================================
# select optimal number of features
thisModel = LogisticRegression(penalty='l1', C=1)
rfecv = RFECV(estimator=thisModel,
step=1,
cv=StratifiedKFold(y, n_folds=3),
scoring='f1')
Xt = rfecv.fit_transform(X, y)
optimalNumberOfFeatures = rfecv.n_features_
introReport = [
'Optimal Number of Attributes: ' + str(optimalNumberOfFeatures),
'The following attributes are the most influential to the outcome'
]
#=======================================================================
# plot number of selected features VS cross-validation scores
plt.figure(figsize=(12, 8))
plt.xlabel("Number of Attributes", fontsize=20)
plt.ylabel("Score", fontsize=20)
plt.title("Attribute Selection", fontsize=25)
plt.plot(range(1, len(rfecv.grid_scores_) + 1), rfecv.grid_scores_)
imgOne = 'static/thePlot.jpg'
plt.savefig('flask_files/' + imgOne, dpi=300)
#=======================================================================
# get the feature feature importance rankings
model = RandomForestClassifier(n_estimators=300)
model.fit(X, y)
theImportances = list(model.feature_importances_)
sortedImportances = sorted(theImportances, reverse=True)
# ...and print the selected features along with their weights and ranks
tableOne = []
for ii in range(1, optimalNumberOfFeatures + 1):
tableOne.append(
dict(Feature=str(theFeatures[theImportances.index(
sortedImportances[ii - 1])]),
Weight=str(sortedImportances[ii - 1]),
Rank=str(ii)))
#=======================================================================
# plot histogram of the most important feature
thisFeature = 0
allThoseFeatures = dp[theFeatures[theImportances.index(
sortedImportances[thisFeature])]]
plt.figure(figsize=(12, 8))
combinedOutcomes = plt.hist(allThoseFeatures, bins=10)
# plt.hist(allThoseFeatures, bins=10)
plt.xlabel('Attribute: ' +
theFeatures[theImportances.index(sortedImportances[0])],
fontsize=20)
plt.ylabel('Count', fontsize=20)
plt.title('Impact of the Most Influential Attribute', fontsize=25)
imgTwo = 'static/theHist.jpg'
plt.savefig('flask_files/' + imgTwo, dpi=300)
#=======================================================================
# plot impact of the most important feature
positiv = allThoseFeatures[yOrig == 1]
negativ = allThoseFeatures[yOrig == 0]
plt.figure(figsize=(12, 8))
negA = plt.hist(negativ, bins=combinedOutcomes[1])
posA = plt.hist(positiv, bins=combinedOutcomes[1])
# yUpperLimit = np.max([negA[0], posA[0]])*1.01
# plt.subplot(1,2,1)
# plt.hist(negativ,bins=combinedOutcomes[1])
# plt.ylim(ymax = yUpperLimit*1.01, ymin = 0)
# plt.xlabel(theFeatures[theImportances.index(sortedImportances[thisFeature])], fontsize=16)
# plt.ylabel('Count', fontsize=16)
# plt.title('Negative', fontsize=20)
#
# plt.subplot(1,2,2)
# plt.hist(positiv,bins=combinedOutcomes[1])
# plt.ylim(ymax = yUpperLimit, ymin = 0)
# plt.xlabel(theFeatures[theImportances.index(sortedImportances[thisFeature])], fontsize=16)
# plt.title('Positive',fontsize=20)
#
# imgThree = 'static/theNegPosHist.jpg'
# plt.savefig('flask_files/'+imgThree, dpi=300)
#=======================================================================
a = posA[0]
b = negA[0]
c = combinedOutcomes[0]
posImpact = np.divide(a, c)
negImpact = np.divide(b, c)
midPoints = []
for i in range(1, len(combinedOutcomes[1])):
midPoints.append(
(combinedOutcomes[1][i] + combinedOutcomes[1][i - 1]) / 2.)
for i in range(len(posImpact)):
if np.isnan(posImpact[i]):
posImpact[i] = 0
if np.isnan(negImpact[i]):
negImpact[i] = 0
plt.figure(figsize=(12, 8))
plt.hold(True)
plt.plot(midPoints, posImpact, '.', markersize=20, label='Positive')
plt.plot(midPoints, negImpact, 'r.', markersize=20, label='Negative')
plt.legend(prop={'size': 20})
plt.xlabel(theFeatures[theImportances.index(
sortedImportances[thisFeature])],
fontsize=16)
plt.ylabel('Relative Impact', fontsize=20)
plt.grid()
imgThree = 'static/theNegPosHist.jpg'
plt.savefig('flask_files/' + imgThree, dpi=300)
#=======================================================================
# generate plots for report (this is save to an "html" file)
from bokeh.charts import Histogram, output_file, show, save, gridplot
from bokeh.plotting import figure
plotList = []
for i in range(optimalNumberOfFeatures):
thisFeatureIs = theFeatures[theImportances.index(sortedImportances[i])]
allThoseFeatures = dp[thisFeatureIs]
combinedOutcomes = plt.hist(allThoseFeatures, bins=10)
positiv = allThoseFeatures[yOrig == 1]
negativ = allThoseFeatures[yOrig == 0]
negA = plt.hist(negativ, bins=combinedOutcomes[1])
posA = plt.hist(positiv, bins=combinedOutcomes[1])
posImpact = np.divide(posA[0], combinedOutcomes[0])
negImpact = np.divide(negA[0], combinedOutcomes[0])
midPoints = []
for i in range(1, len(combinedOutcomes[1])):
midPoints.append(
(combinedOutcomes[1][i] + combinedOutcomes[1][i - 1]) / 2.)
for i in range(len(posImpact)):
if np.isnan(posImpact[i]):
posImpact[i] = 0
if np.isnan(negImpact[i]):
negImpact[i] = 0
hist0 = Histogram(dp,
values=thisFeatureIs,
color='blue',
title="Impact of " + thisFeatureIs,
bins=10)
plot0 = figure()
plot0.xaxis.axis_label = thisFeatureIs
plot0.yaxis.axis_label = "Relative Impact"
# plot0.title = "Relative Impact of " + thisFeatureIs
plot0.circle(midPoints,
list(negImpact),
size=10,
color="red",
alpha=0.9,
legend='Negative')
plot0.circle(midPoints,
list(posImpact),
size=10,
color="green",
alpha=0.9,
legend='Positive')
plotList.append([hist0, plot0])
output_file("flask_files/static/Report.html", title="Report")
hist = gridplot(plotList)
save(hist)
#=======================================================================
# specify the models to run tests with
theModels = {
'Logistic Regression': LogisticRegression(penalty='l1'),
'LDA': LinearDiscriminantAnalysis(),
'SVM': SVC(kernel='linear'),
'Random Forest': RandomForestClassifier(n_estimators=300)
}
# ...then display the results of the tests
classifierComparisons = []
for aModel in theModels:
model = theModels[aModel]
results = cross_validation.cross_val_score(model,
Xt,
y,
scoring='f1',
cv=StratifiedKFold(
y, n_folds=3))
classifierComparisons.append(
dict(Classifier=aModel, Score=np.max(results)))
#=======================================================================
# display the plots
theJPGs = [imgOne, imgTwo, imgThree]
#=======================================================================
return introReport, tableOne, optimalNumberOfFeatures, classifierComparisons, theJPGs
def preProcess(theFileName):
df = pd.read_csv(str(theFileName))
if 'Unnamed: 0' in df.columns:
df = df.drop('Unnamed: 0', axis = 1)
labBin = sklearn.preprocessing.LabelBinarizer()
df['y'] = labBin.fit_transform(df['y'])
dp = pd.get_dummies(df)
X = dp.drop('y', axis = 1)
y = dp[['y']]
# get the features
theFeatures = X.columns
# convert the dataframes to arrays
X = X.values
y = y.values
y.shape = np.shape(y)[0]
yOrig = y[:] # need this later for plotting feature impacts
# and carry out feature scaling
X = StandardScaler().fit_transform(X)
#=======================================================================
# apply random undersampling if labels are imbalanced
labelSkewness = 100*np.sum(y)*1./np.shape(y)[0]
if np.min([labelSkewness, 100-labelSkewness]) < (100./3.):
rus = RandomUnderSampler(verbose=0)
X, y = rus.fit_sample(X, y)
#=======================================================================
# select optimal number of features
thisModel = LogisticRegression(penalty='l1', C=1)
rfecv = RFECV(estimator=thisModel, step=1, cv=StratifiedKFold(y, n_folds=3), scoring='f1')
Xt = rfecv.fit_transform(X, y);
optimalNumberOfFeatures = rfecv.n_features_
introReport = ['Optimal Number of Attributes: ' + str(optimalNumberOfFeatures), 'The following attributes are the most influential to the outcome']
#=======================================================================
# plot number of selected features VS cross-validation scores
plt.figure(figsize=(12, 8))
plt.xlabel("Number of Attributes", fontsize=20)
plt.ylabel("Score", fontsize=20)
plt.title("Attribute Selection", fontsize=25)
plt.plot(range(1, len(rfecv.grid_scores_) + 1), rfecv.grid_scores_)
imgOne = 'static/thePlot.jpg'
plt.savefig('flask_files/'+imgOne, dpi=300)
#=======================================================================
# get the feature feature importance rankings
model = RandomForestClassifier(n_estimators=300)
model.fit(X,y)
theImportances = list(model.feature_importances_)
sortedImportances = sorted(theImportances,reverse = True)
# ...and print the selected features along with their weights and ranks
tableOne = []
for ii in range(1,optimalNumberOfFeatures+1):
tableOne.append(dict(Feature = str(theFeatures[theImportances.index(sortedImportances[ii-1])]), Weight = str(sortedImportances[ii-1]), Rank = str(ii)))
#=======================================================================
# plot histogram of the most important feature
thisFeature = 0
allThoseFeatures = dp[theFeatures[theImportances.index(sortedImportances[thisFeature])]]
plt.figure(figsize=(12, 8))
combinedOutcomes = plt.hist(allThoseFeatures, bins=10)
# plt.hist(allThoseFeatures, bins=10)
plt.xlabel('Attribute: ' + theFeatures[theImportances.index(sortedImportances[0])], fontsize=20)
plt.ylabel('Count', fontsize=20)
plt.title('Impact of the Most Influential Attribute', fontsize=25)
imgTwo = 'static/theHist.jpg'
plt.savefig('flask_files/'+imgTwo, dpi=300)
#=======================================================================
# plot impact of the most important feature
positiv = allThoseFeatures[yOrig==1]
negativ = allThoseFeatures[yOrig==0]
plt.figure(figsize=(12, 8))
negA = plt.hist(negativ,bins=combinedOutcomes[1])
posA = plt.hist(positiv,bins=combinedOutcomes[1])
# yUpperLimit = np.max([negA[0], posA[0]])*1.01
# plt.subplot(1,2,1)
# plt.hist(negativ,bins=combinedOutcomes[1])
# plt.ylim(ymax = yUpperLimit*1.01, ymin = 0)
# plt.xlabel(theFeatures[theImportances.index(sortedImportances[thisFeature])], fontsize=16)
# plt.ylabel('Count', fontsize=16)
# plt.title('Negative', fontsize=20)
#
# plt.subplot(1,2,2)
# plt.hist(positiv,bins=combinedOutcomes[1])
# plt.ylim(ymax = yUpperLimit, ymin = 0)
# plt.xlabel(theFeatures[theImportances.index(sortedImportances[thisFeature])], fontsize=16)
# plt.title('Positive',fontsize=20)
#
# imgThree = 'static/theNegPosHist.jpg'
# plt.savefig('flask_files/'+imgThree, dpi=300)
#=======================================================================
a = posA[0]
b = negA[0]
c = combinedOutcomes[0]
posImpact = np.divide(a,c)
negImpact = np.divide(b,c)
midPoints=[]
for i in range(1,len(combinedOutcomes[1])):
midPoints.append((combinedOutcomes[1][i] + combinedOutcomes[1][i-1])/2.)
for i in range(len(posImpact)):
if np.isnan(posImpact[i]):
posImpact[i]=0
if np.isnan(negImpact[i]):
negImpact[i]=0
plt.figure(figsize=(12, 8))
plt.hold(True)
plt.plot(midPoints, posImpact,'.', markersize=20, label='Positive')
plt.plot(midPoints, negImpact, 'r.', markersize=20, label='Negative')
plt.legend(prop={'size':20})
plt.xlabel(theFeatures[theImportances.index(sortedImportances[thisFeature])], fontsize=16)
plt.ylabel('Relative Impact', fontsize=20)
plt.grid()
imgThree = 'static/theNegPosHist.jpg'
plt.savefig('flask_files/'+imgThree, dpi=300)
#=======================================================================
# generate plots for report (this is save to an "html" file)
from bokeh.charts import Histogram, output_file, show, save, gridplot
from bokeh.plotting import figure
plotList=[]
for i in range(optimalNumberOfFeatures):
thisFeatureIs = theFeatures[theImportances.index(sortedImportances[i])]
allThoseFeatures = dp[thisFeatureIs]
combinedOutcomes = plt.hist(allThoseFeatures, bins=10)
positiv = allThoseFeatures[yOrig==1]
negativ = allThoseFeatures[yOrig==0]
negA = plt.hist(negativ,bins=combinedOutcomes[1])
posA = plt.hist(positiv,bins=combinedOutcomes[1])
posImpact = np.divide(posA[0],combinedOutcomes[0])
negImpact = np.divide(negA[0],combinedOutcomes[0])
midPoints=[]
for i in range(1,len(combinedOutcomes[1])):
midPoints.append((combinedOutcomes[1][i] + combinedOutcomes[1][i-1])/2.)
for i in range(len(posImpact)):
if np.isnan(posImpact[i]):
posImpact[i]=0
if np.isnan(negImpact[i]):
negImpact[i]=0
hist0 = Histogram(dp, values=thisFeatureIs, color='blue', title="Impact of " + thisFeatureIs, bins=10)
plot0 = figure()
plot0.xaxis.axis_label = thisFeatureIs
plot0.yaxis.axis_label = "Relative Impact"
# plot0.title = "Relative Impact of " + thisFeatureIs
plot0.circle(midPoints, list(negImpact), size=10, color="red", alpha=0.9, legend='Negative')
plot0.circle(midPoints, list(posImpact), size=10, color="green", alpha=0.9, legend='Positive')
plotList.append([hist0,plot0])
output_file("flask_files/static/Report.html", title = "Report")
hist = gridplot(plotList)
save(hist)
#=======================================================================
# specify the models to run tests with
theModels = {'Logistic Regression':LogisticRegression(penalty='l1'), 'LDA':LinearDiscriminantAnalysis(), 'SVM':SVC(kernel='linear'), 'Random Forest':RandomForestClassifier(n_estimators=300)}
# ...then display the results of the tests
classifierComparisons=[]
for aModel in theModels:
model = theModels[aModel]
results = cross_validation.cross_val_score(model, Xt, y, scoring='f1', cv=StratifiedKFold(y, n_folds=3))
classifierComparisons.append(dict(Classifier = aModel, Score = np.max(results)))
#=======================================================================
# display the plots
theJPGs = [imgOne, imgTwo, imgThree]
#=======================================================================
return introReport, tableOne, optimalNumberOfFeatures, classifierComparisons, theJPGs
from unbalanced_dataset.under_sampling import RandomUnderSampler
# Generate the dataset
X, y = make_classification(n_classes=2, class_sep=2, weights=[0.1, 0.9],
n_informative=3, n_redundant=1, flip_y=0,
n_features=20, n_clusters_per_class=1,
n_samples=5000, random_state=10)
# Instanciate a PCA object for the sake of easy visualisation
pca = PCA(n_components=2)
# Fit and transform x to visualise inside a 2D feature space
X_vis = pca.fit_transform(X)
# Apply the random under-sampling
rus = RandomUnderSampler()
X_resampled, y_resampled = rus.fit_transform(X, y)
X_res_vis = pca.transform(X_resampled)
# Two subplots, unpack the axes array immediately
f, (ax1, ax2) = plt.subplots(1, 2)
ax1.scatter(X_vis[y == 0, 0], X_vis[y == 0, 1], label="Class #0", alpha=0.5,
edgecolor=almost_black, facecolor=palette[0], linewidth=0.15)
ax1.scatter(X_vis[y == 1, 0], X_vis[y == 1, 1], label="Class #1", alpha=0.5,
edgecolor=almost_black, facecolor=palette[2], linewidth=0.15)
ax1.set_title('Original set')
ax2.scatter(X_res_vis[y_resampled == 0, 0], X_res_vis[y_resampled == 0, 1],
label="Class #0", alpha=.5, edgecolor=almost_black,
facecolor=palette[0], linewidth=0.15)