## outfile.close()
## sys.exit(0)
normImage = False
makePlot = False
alldata = cPickle.load( file('data/alldata_'+('norm_' if normImage else '')+ 'TRAIN.pkl', 'r') )
print "### Building Model ###"
X = alldata[:,1:]
y = alldata[:,0]
bfish = BinnedFisher( bins=[0.25, 0.5, 0.75, float('inf')] )
#bfish.fit(X, y, tol=[4.0e-3, 1.0e-3, 0.5e-3]) #old
if normImage:
#bfish.fit(X, y, tol=[2.0e-3, 0.6e-3, 0.2e-3]) #good for normed, 10k per bin per label
bfish.fit(X, y, tol=[1.0e-3, 0.75e-4, 0.1e-3]) #good for normed, 10k per bin per label
else:
#bfish.fit(X, y, tol=[9.5e0, 11e0, 3.0e0]) #good-ish for non-normed, 10k per bin per label
bfish.fit(X, y, tol=[2.5e0, 2.5e-1, 0.3e0]) #good-ish for non-normed, 10k per bin per label
outfile = file('trained_'+('norm_' if normImage else '')+ 'DR_fisher.pkl', 'wb')
cPickle.dump(bfish, outfile, protocol=cPickle.HIGHEST_PROTOCOL)
outfile.close()
X = np.vstack((class0, class1))
y = np.array([0 for i in range(50)] + [1 for i in range(50)])
f = Fisher()
f.fit(X, y, tol=0.1)
print f.transform(X)
# after fit, can update tolerance
f.update_tol(tol=0.01)
print f.transform(X)
# add additional variable to binning, for BinnedFisher
v = np.array(
[[0.25 for i in range(25)] + [0.75 for i in range(25)] + [0.25 for i in range(25)] + [0.75 for i in range(25)]]
)
X = np.hstack((v.T, X))
bf = BinnedFisher(bins=[0.0, 0.5, 1.0])
bf.fit(X, y, tol=[0.01, 0.01])
print bf.transform(X)
## outfile.close()
## sys.exit(0)
normImage = False
makePlot = False
alldata = cPickle.load(
file('data/alldata_' + ('norm_' if normImage else '') + 'TRAIN.pkl', 'r'))
print "### Building Model ###"
X = alldata[:, 1:]
y = alldata[:, 0]
bfish = BinnedFisher(bins=[0.25, 0.5, 0.75, float('inf')])
#bfish.fit(X, y, tol=[4.0e-3, 1.0e-3, 0.5e-3]) #old
if normImage:
#bfish.fit(X, y, tol=[2.0e-3, 0.6e-3, 0.2e-3]) #good for normed, 10k per bin per label
bfish.fit(X, y, tol=[1.0e-3, 0.75e-4,
0.1e-3]) #good for normed, 10k per bin per label
else:
#bfish.fit(X, y, tol=[9.5e0, 11e0, 3.0e0]) #good-ish for non-normed, 10k per bin per label
bfish.fit(X, y,
tol=[2.5e0, 2.5e-1,
0.3e0]) #good-ish for non-normed, 10k per bin per label
outfile = file('trained_' + ('norm_' if normImage else '') + 'DR_fisher.pkl',
'wb')
#make random two classes
class0 = np.random.normal(1, 1, (50, 5))
class1 = np.random.normal(-2, 0.5, (50, 5))
X = np.vstack((class0, class1))
y = np.array([0 for i in range(50)] + [1 for i in range(50)])
f = Fisher()
f.fit(X, y, tol=0.1)
print f.transform(X)
#after fit, can update tolerance
f.update_tol(tol=0.01)
print f.transform(X)
#add additional variable to binning, for BinnedFisher
v = np.array([[0.25 for i in range(25)] + [0.75 for i in range(25)] +
[0.25 for i in range(25)] + [0.75 for i in range(25)]])
X = np.hstack((v.T, X))
bf = BinnedFisher(bins=[0.0, 0.5, 1.0])
bf.fit(X, y, tol=[0.01, 0.01])
print bf.transform(X)
