nugget1.py

#!/usr/bin/env python

import numpy as np
from scikits.learn import linear_model
from dldata import hongmajaj  # Hong & Majaj Neuronal Datasets module

# -- random number generator
rng = np.random.RandomState(42)

# -- dataset object
ds = hongmajaj.dataset.ITChaboTitoVar06wh(verify_sha1=True)

# -- get data
time_window = (70, 170)
print 'getting X'
xl = [ds.get_neurons(imgd, time_window)
     for imgd in ds.meta
     if 'blank' not in imgd['annotations']['labels']]
X = np.array(xl)

print 'getting y'
yl = [imgd['annotations']['pose']['ty']  # translation on the y axis
      for imgd in ds.meta
      if 'blank' not in imgd['annotations']['labels']]
y = np.array(yl)

# -- shuffle data
ridx = rng.permutation(len(y))
X = X[ridx]
y = y[ridx]

# training split
X_trn = X[::2]
y_trn = y[::2]

# testing split
X_tst = X[1::2]
y_tst = y[1::2]

# -- training
print 'fit'
clf = linear_model.LassoCV()
clf.fit(X_trn, y_trn)

# -- testing
print 'predict'
y_gv = clf.predict(X_tst)

cc = np.corrcoef(y_gv, y_tst)[0, 1]

# -- plot
import pylab as pl
pl.scatter(y_tst, y_gv)
pl.xlabel('translation y (ground truth)')
pl.ylabel('translation y (predicted)')
pl.title('hongmajaj.ITChaboTitoVar06wh -- half/half -- Lasso Regression (CC=%0.2f)' % cc)
pl.savefig('nugget1.png')