def find_best_features(year, features, sex, age, heavy):
"""year=-1 => both years 2,3 """
print 'find_best_features(year=%d,features=%s,sex=%s,age=%s,heavy=%s)' % (year, features, sex,
age, heavy)
X, y, keys = getXy_by_features(year, features, sex, age)
title = 'features=%s,sex=%s,age=%s,year=%d' % (features,sex,age,year)
results, n_samples = select_features.get_most_predictive_feature_set(title, X, y, keys, heavy)
return results, n_samples, keys
def find_best_features(year, features, sex):
import select_features
print 'find_best_features(year=%d)' % year
if features == 'pcg':
X,y,keys = getXy_pcg(year)
elif features == 'patient':
X,y,keys = getXy_patient(year)
elif features == 'all':
X,y,keys = getXy_all(year)
print 'keys=%s' % keys
if sex and sex.lower()[0] in 'mf' and 'Sex' in keys:
# Get male or female population
sex_key = keys.index('Sex')
if sex.lower()[0] == 'm':
p = X[:,sex_key] < 0.5
else:
p = X[:,sex_key] > 0.5
X = X[p,:]
y = y[p]
# Remove columns with low counts
LOW_COUNT_THRESHOLD = 100
Xtot = X.sum(axis=0)
significant = Xtot >= LOW_COUNT_THRESHOLD
# Remove sex too
significant[sex_key] = False
print 'Removing keys < %d: %s' % (LOW_COUNT_THRESHOLD,
[keys[i] for i in range(len(keys)) if not significant[i]])
print 'keys=%d X=%s => ' % (len(keys), X.shape),
keys = [keys[i] for i in range(len(keys)) if significant[i]]
X = X[:,significant]
print 'keys=%d X=%s' % (len(keys), X.shape)
# Normalize
means = X.mean(axis=0)
stds = X.std(axis=0)
for i in range(X.shape[1]):
X[:,i] = X[:,i] - means[i]
if abs(stds[i]) > 1e-6:
X[:,i] = X[:,i]/stds[i]
return select_features.get_most_predictive_feature_set(X, y, keys), keys
