def run():
print(' => Reading features dataset')
df = pd.read_csv('results/processed_extracted_features.csv').drop('ids',
axis=1)
y = df['Diagnosis'].values
x = df[df.columns[1:]].values
x = StandardScaler().fit_transform(x)
results = pd.DataFrame(
columns=['Classifier', 'Params', 'BAS', 'BER', 'MCC', 'Accuracy'])
print(' => Testing classifiers')
print(' ==> Naive Bayes .. ', end='\r')
evaluation = evaluate(GaussianNB, x, y)
evaluation.update({'Classifier': 'Naive Bayes'})
results = results.append(evaluation, ignore_index=True).round(4)
print(f' ==> Naive Bayes .. {evaluation["BAS"]}')
print(' ==> Random Forest .. ', end='\r')
evaluation = evaluate(RandomForestClassifier,
x,
y,
params={'n_estimators': 50})
evaluation.update({'Classifier': 'Random Forest', 'Params': '# Trees: 50'})
results = results.append(evaluation, ignore_index=True).round(4)
print(f' ==> Random Forest .. {evaluation["BAS"]}')
print(' ==> Ada Boost .. ', end='\r')
evaluation = evaluate(AdaBoostClassifier,
x,
y,
params={'learning_rate': 0.05})
evaluation.update({
'Classifier': 'Ada Boost',
'Params': 'Learning rate: 0.5'
})
results = results.append(evaluation, ignore_index=True).round(4)
print(f' ==> Ada Boost .. {evaluation["BAS"]}')
print(' ==> SVC .. ', end='\r')
evaluation = evaluate(SVC, x, y, params={'kernel': 'linear', 'C': 0.1})
evaluation.update({
'Classifier': 'SVM',
'Params': 'kernel: linear, C: 0.1'
})
results = results.append(evaluation, ignore_index=True).round(4)
print(f' ==> SVC .. {evaluation["BAS"]}')
print(' ==> KNN .. ', end='\r')
evaluation = evaluate(KNeighborsClassifier,
x,
y,
params={'n_neighbors': 7})
evaluation.update({'Classifier': 'KNN', 'Params': 'K: 7'})
results = results.append(evaluation, ignore_index=True).round(4)
print(f' ==> KNN .. {evaluation["BAS"]}')
print(' => Done!')
return results
def ffs():
stdout.write(' => Reading DF')
df = pd.read_csv(TRAIN_URI)
stdout.write('\r => Gettting FDR ')
fdr = FDR(df)
stdout.write('\r => Initializing sets ')
ir = set(['CDR'])
not_ir = set(df.columns[1:])
not_ir.remove('CDR')
Y = df['Diagnosis']
res = pd.DataFrame(columns=['feature', 'Accuracy'])
evaluation = evaluate(SVC,
df[['CDR']].values,
Y,
params={
'C': 0.1,
'kernel': 'linear'
})
evaluation['feature'] = 'CDR'
res = res.append(evaluation, ignore_index=True)
for i in range(len(not_ir)):
features = list(ir)
best_result = {SELECTION_CRITERIA: 0}
for f in not_ir:
values = df[features + [f]].values
evaluation = evaluate(SVC,
values,
Y,
params={
'C': 0.1,
'kernel': 'linear'
})
if evaluation[SELECTION_CRITERIA] > best_result[SELECTION_CRITERIA]:
best_result = evaluation
best_result['feature'] = f
elif evaluation[SELECTION_CRITERIA] == best_result[
SELECTION_CRITERIA]:
champion = best_result['feature']
challenger = f
if fdr[challenger] > fdr[champion]:
best_result = evaluation
best_result['feature'] = f
res = res.append(best_result, ignore_index=True)
not_ir.remove(best_result['feature'])
ir.add(best_result['feature'])
stdout.write('\r ==> %d features selected .. %0.04f sensibility' %
(res.shape[0], best_result[SELECTION_CRITERIA]))
stdout.write('\n')
return res
def grid_search():
"""Main handler.
This function looks for the best classifier & it's best combination of
parameters.
"""
cols = ['Accuracy', 'BAS', 'BER', 'MCC', 'Sensibility', 'Specificity']
all_res = pd.DataFrame(columns=(['Classifier'] + cols))
print(' => Reading features dataset')
df = pd.read_csv(TRAIN_URI) # .drop('ids', axis=1)
y = df['Diagnosis'].values
x = df[df.columns[1:]].values
print('\nNaive Bayes')
results = pd.DataFrame([evaluate(GaussianNB, x, y)], columns=cols)
print(results)
results['Classifier'] = 'Naive Bayes'
all_res = all_res.append(results.iloc[0], ignore_index=True)
print('\nRandom Forest')
results = pd.DataFrame(
[evaluate(RandomForestClassifier, x, y, params={'n_estimators': 50})],
columns=cols)
print(results)
results['Classifier'] = 'Random Forest'
all_res = all_res.append(results.iloc[0], ignore_index=True)
print('\nAda Boost')
results = ada_boost.grid_search(x, y).sort_values('Accuracy',
ascending=False)
print(results)
results['Classifier'] = 'Ada Boost'
all_res = all_res.append(results.iloc[0], ignore_index=True)
print('\nSVC linear')
results = linear_svm.grid_search(x, y).sort_values('Accuracy',
ascending=False)
print(results)
results['Classifier'] = 'SVC'
all_res = all_res.append(results.iloc[0], ignore_index=True)
print('\nKNN')
results = knn.grid_search(x, y).sort_values('Accuracy', ascending=False)
print(results)
results['Classifier'] = 'KNN'
all_res = all_res.append(results.iloc[0], ignore_index=True)
print('\n', all_res.sort_values('Accuracy', ascending=False))
def grid_search(values, target, verbose=True):
"""Looks for the best param combinations for SVC."""
stdout.write(' => Best Cs for SVM\n')
results = pd.DataFrame(columns=['C', 'Accuracy'])
cs = [10**c for c in range(-3, 2)]
for i, c in enumerate(cs):
if verbose:
stdout.write(f'\r ==> SVC .... {i + 1}/{len(cs)}')
evaluation = evaluate(SVC,
values,
target,
params={
'C': c,
'kernel': 'linear'
})
evaluation.update({'C': c})
results = results.append(evaluation, ignore_index=True).round(4)
if verbose:
stdout.write(
f'\x1b[2k\r => Best SVM, {results.shape[0]} combs tested!\n')
return results
def get_best_comb(images, target, msg=''):
"""Looks for the best distance on Haralick features."""
results = pd.DataFrame()
# In disc
for dist in range(1, 4):
print(f' ==> ({msg}) Distance {dist}/3 (all degrees + mean)', end='\r')
haralick = extractor.get_haralick(images, dist, HARALICK_NAMES)
values = StandardScaler().fit_transform(haralick)
res = evaluator.evaluate(GaussianNB, values, target)
res.update({'Distance': dist})
results = results.append(res, ignore_index=True).round(4)
return results
def grid_search(values, target, verbose=True):
"""Looks for the best param combinations for KNN."""
stdout.write(' => Best Ks for KNN\n')
results = pd.DataFrame(columns=['K', 'Accuracy'])
ks = [k for k in range(1, 20, 2)]
for i, k in enumerate(ks):
if verbose:
stdout.write(f'\r ==> KNN .... {i + 1}/{len(ks)}')
evaluation = evaluate(KNeighborsClassifier,
values,
target,
params={'n_neighbors': k})
evaluation.update({'K': k})
results = results.append(evaluation, ignore_index=True).round(4)
if verbose:
stdout.write(f'\x1b[2k\r => KNN, {results.shape[0]} combs tested!\n')
return results
def grid_search(values, target, verbose=True):
"""Looks for the best param combinations for AdaBoost."""
if verbose:
stdout.write(' => Getting best LRates for AdaBoost\n')
results = pd.DataFrame(columns=['Learning rate', 'Accuracy'])
l_rates = [lr / 1000 for lr in range(80, 121, 10)]
for i, lr in enumerate(l_rates):
if verbose:
stdout.write(f'\r ==> AdaBoost LRates .... {i + 1}/{len(l_rates)}')
evaluation = evaluate(AdaBoostClassifier,
values,
target,
params={'learning_rate': lr})
evaluation.update({'Learning rate': lr})
results = results.append(evaluation, ignore_index=True).round(4)
if verbose:
stdout.write(
f'\x1b[2k\r => AdaBoost, {results.shape[0]} combs tested! \n')
return results
from lib.evaluator import evaluate
from lib import rimone
POINTS = [6, 7, 8, 9, 10]
RADIUS = [1, 2, 3]
ds = rimone.dataset()
Y = ds.Y
res = pd.DataFrame()
for p in POINTS:
for r in RADIUS:
print(f'p: {p}, r: {r}')
X = StandardScaler().fit_transform(get_lbp(ds.cups, radius=r, points=p))
evaluation = evaluate(GaussianNB, X, Y)
evaluation.update({
'radius': r,
'point': p
})
res = res.append(evaluation, ignore_index=True).sort_values('Score', ascending=False)
res.to_csv('results/lbp_cups.csv', index=False)
print(res.sort_values('Score', ascending=False))
POINTS = [6, 7, 8, 9, 10]
RADIUS = [1, 2, 3]
res = pd.DataFrame()
for p in POINTS: