47 lines
3.6 KiB
Python
47 lines
3.6 KiB
Python
import unittest
|
|
import numpy as np
|
|
from phase2_e2_statistics import weights_for_components,zero_support,adjusted_estimate,interval,raw_gate,analyse
|
|
|
|
class Statistics(unittest.TestCase):
|
|
def test_component_weights_and_resample(self):
|
|
w,ids,g=weights_for_components(['a','a','b']);np.testing.assert_allclose(w,[.25,.25,.5])
|
|
w,_,_=weights_for_components(['a','a','b'],[2,0]);np.testing.assert_allclose(w,[.5,.5,0])
|
|
def test_no_subset_renormalization(self):
|
|
x=np.array([[-1,0],[1,0],[np.nan,np.nan]]);d=np.array([-1.,3.,10.]);w=np.array([.25,.25,.5]);v=adjusted_estimate(d,x,[True,True,False],w)
|
|
self.assertAlmostEqual(v['value'],5.5);self.assertAlmostEqual(v['alpha'],1);self.assertEqual(v['dual_valid_weight'],.5)
|
|
def test_uncentered_correction(self):
|
|
x=np.array([[-1,-1],[1,-1],[-1,1],[1,1],[.5,.5]]);d=2+3*x[:,0]-4*x[:,1]
|
|
v=adjusted_estimate(d,x,np.ones(5,bool),np.ones(5)/5);self.assertAlmostEqual(v['value'],2);self.assertAlmostEqual(v['alpha'],2)
|
|
def test_zero_column_and_collinear_estimability(self):
|
|
for x in [np.zeros((3,2)),np.array([[-1,-2],[0,0],[1,2]])]:
|
|
v=adjusted_estimate(np.ones(3),x,[True]*3,np.ones(3)/3);self.assertTrue(v['zero_point_estimable']);self.assertAlmostEqual(v['value'],1)
|
|
def test_rank_deficient_offset_and_outside_hull(self):
|
|
x=np.array([[-1,1],[0,1],[1,1]]);v=adjusted_estimate(np.ones(3),x,[True]*3,np.ones(3)/3);self.assertIsNone(v['value']);self.assertFalse(v['zero_point_estimable'])
|
|
x=np.array([[1,1],[2,1],[1,2]]);v=adjusted_estimate(np.ones(3),x,[True]*3,np.ones(3)/3);self.assertTrue(v['zero_point_estimable']);self.assertIsNone(v['value'])
|
|
def test_origin_on_boundary_and_no_support(self):
|
|
self.assertTrue(zero_support([[0,0],[1,0],[0,1]])['supported']);self.assertFalse(zero_support([[2,0],[3,0]])['supported']);self.assertFalse(zero_support([])['supported'])
|
|
def test_exact_five_percent_boundary(self):
|
|
self.assertFalse(interval([np.nan]*5+[1]*95,1,100)['usable']);self.assertTrue(interval([np.nan]*4+[1]*96,1,100)['usable'])
|
|
def test_raw_gate_strict_and_complete(self):
|
|
f=dict(high=-.001,defined_components=204);l=dict(high=.1,defined_components=204)
|
|
self.assertEqual(raw_gate(f,l),'EXPAND_FIXED_TYPE_CONTEXT');self.assertEqual(raw_gate(dict(f,high=0),l),'PRUNE_FIXED_TYPE_CONTEXT_RECIPE');self.assertEqual(raw_gate(dict(f,defined_components=203),l),'PARK_UNDEFINED')
|
|
self.assertEqual(raw_gate(dict(f,high=.001),dict(l,high=-.1)),'PRUNE_FIXED_TYPE_CONTEXT_RECIPE')
|
|
def test_bootstrap_refits_and_retains_components(self):
|
|
rows=[];xs=[-1.,0.,1.,2.,-2.,.5];ds=[-1.,2.,1.,5.,0.,3.]
|
|
for i,(x,d) in enumerate(zip(xs,ds)):
|
|
geom={a:dict(valid=True,descriptor=dict(cut_area=float(np.exp(x)) if a=='T' else 1.,rms_diameter=1.)) for a in ['T','C','G']}
|
|
metrics={a:{m:dict(value=(d if a=='T' else 0.)) for m in ['relative_l2','last_location','kendall_tau_b']} for a in ['T','C','G']}
|
|
rows.append(dict(case=str(i),component_id=str(i),metrics=metrics,geometry=dict(arms=geom)))
|
|
result,arr=analyse(rows,replicates=40,seed=3,expected_parts=6,expected_components=6)
|
|
x=np.column_stack([xs,np.zeros(6)]);w=np.ones(6)/6;point=adjusted_estimate(ds,x,[True]*6,w);different=0
|
|
for k,draw in enumerate(arr['component_draws']):
|
|
wk=np.bincount(draw,minlength=6)/6;v=adjusted_estimate(ds,x,[True]*6,wk);got=arr['T-G_relative_l2_adjusted'][k]
|
|
if v['value'] is None:self.assertTrue(np.isnan(got))
|
|
else:
|
|
self.assertAlmostEqual(got,v['value']);fixed=np.dot(wk,ds)-np.sum(wk*(x@np.array(point['beta_minimum_norm'])))
|
|
different+=abs(got-fixed)>1e-8
|
|
self.assertGreater(different,0)
|
|
with self.assertRaises(ValueError):analyse(rows,replicates=2)
|
|
|
|
if __name__=='__main__':unittest.main()
|