def test_create_cache_train_once(self):
self.ctx = context.DataContext(store.MemoryStore('test', verbose=True),
self.data, train_once=True)
f = base.Normalize(base.F(10) + base.F('a'))
ctx = self.ctx
r = f.create(ctx)
r = r[r.columns[0]]
self.assertAlmostEqual(r.mean(), 0)
self.assertAlmostEqual(r.std(), 1)
# now add some new data
idx = len(self.data) + 1000
ctx.data = ctx.data.append(DataFrame([100, 200], columns=['a'], index=Index([idx, idx+1])))
r = f.create(ctx)
r = r[r.columns[0]]
self.assertAlmostEqual(r[idx], (100 - self.data['a'].mean()) / self.data['a'].std())
# drop all the other data ... should still use old prep data
ctx.data = ctx.data.ix[[idx, idx+1]]
r = f.create(ctx)
r = r[r.columns[0]]
self.assertAlmostEqual(r[idx], (100 - self.data['a'].mean()) / self.data['a'].std())
# new train_index should NOT bust old cache
ctx.data = ctx.data.ix[[idx, idx+1]]
ctx.train_index = ctx.data.index
ctx.prep_index = ctx.data.index
r = f.create(ctx)
r = r[r.columns[0]]
self.assertAlmostEqual(r[idx], (100 - self.data['a'].mean()) / self.data['a'].std())
def setUp(self):
self.data = make_data(10)
self.ctx = context.DataContext(store.MemoryStore('test', verbose=True),
self.data)
def setUp(self):
self.data = make_data(10)
self.data['groups'] = self.data['ints'].apply(lambda x: x > 5)
self.ctx = context.DataContext(store.MemoryStore(verbose=True), self.data)