def significance(ev1, ev2, cutoff):
numpy = grace.get_numpy()
# Avoid duplicates in SIG_CACHE
if ev2 < ev1:
ev1, ev2 = ev2, ev1
options = {}
for item in ev1:
options[item[0]] = len(options)
for item in ev2:
if item[0] not in options:
options[item[0]] = len(options)
n = len(options)
matrix = numpy.zeros((n, 2), int)
for item in ev1:
matrix[options[item[0]], 0] = item[1]
for item in ev2:
matrix[options[item[0]], 1] = item[1]
#s = fexact(matrix)
#if s == 0:
# print matrix, s
#return fexact(matrix)
key = (tuple(int(i) for i in matrix.ravel()), cutoff)
if key not in SIG_CACHE:
SIG_CACHE[key] = fexact(matrix, cutoff)
return SIG_CACHE[key]
def significance(ev1, ev2, cutoff):
numpy = grace.get_numpy()
# Avoid duplicates in SIG_CACHE
if ev2 < ev1:
ev1,ev2 = ev2,ev1
options = { }
for item in ev1:
options[item[0]] = len(options)
for item in ev2:
if item[0] not in options:
options[item[0]] = len(options)
n = len(options)
matrix = numpy.zeros((n,2),int)
for item in ev1:
matrix[ options[item[0]], 0 ] = item[1]
for item in ev2:
matrix[ options[item[0]], 1 ] = item[1]
#s = fexact(matrix)
#if s == 0:
# print matrix, s
#return fexact(matrix)
key = (tuple(int(i) for i in matrix.ravel()),cutoff)
if key not in SIG_CACHE:
SIG_CACHE[key] = fexact(matrix,cutoff)
return SIG_CACHE[key]
def fexact(matrix, significance_cutoff):
numpy = grace.get_numpy()
matrix = numpy.array(matrix)
n_row, n_col = matrix.shape
row_sum = numpy.sum(matrix, 1)
col_sum = numpy.sum(matrix, 0)
n = numpy.sum(row_sum)
cutoff = sum([log_fac(item) for item in matrix.ravel()]) * SUM_ERROR_MARGIN
const_part = sum([log_fac(item) for item in row_sum]) + sum(
[log_fac(item) for item in col_sum]) - log_fac(n)
significance = [0.0]
row_remainders = row_sum.copy()
def generate(row, col, col_remainder, total):
cell_min = max(0, col_remainder - numpy.sum(row_remainders[row + 1:]))
cell_max = min(col_remainder, row_remainders[row])
#next_row_remainders = row_remainders.copy()
old_row_remainder = row_remainders[row]
for i in xrange(cell_min, cell_max + 1):
next_total = total + log_fac(i)
row_remainders[row] = old_row_remainder - i
if row + 1 >= n_row:
if col + 1 >= n_col:
if next_total >= cutoff:
significance[0] += numpy.exp(const_part - next_total)
if significance[0] > significance_cutoff:
raise Cutoff_exceeded
else:
generate(0, col + 1, col_sum[col + 1], next_total)
else:
generate(row + 1, col, col_remainder - i, next_total)
row_remainders[row] = old_row_remainder
try:
generate(0, 0, col_sum[0], 0.0)
except Cutoff_exceeded:
return None
return significance[0]
def fexact(matrix, significance_cutoff):
numpy = grace.get_numpy()
matrix = numpy.array(matrix)
n_row, n_col = matrix.shape
row_sum = numpy.sum(matrix,1)
col_sum = numpy.sum(matrix,0)
n = numpy.sum(row_sum)
cutoff = sum([ log_fac(item) for item in matrix.ravel() ]) * SUM_ERROR_MARGIN
const_part = sum([ log_fac(item) for item in row_sum ]) + sum([ log_fac(item) for item in col_sum ]) - log_fac(n)
significance = [ 0.0 ]
row_remainders = row_sum.copy()
def generate(row, col, col_remainder, total):
cell_min = max(0, col_remainder - numpy.sum(row_remainders[row+1:]))
cell_max = min(col_remainder, row_remainders[row])
#next_row_remainders = row_remainders.copy()
old_row_remainder = row_remainders[row]
for i in xrange(cell_min,cell_max+1):
next_total = total + log_fac(i)
row_remainders[row] = old_row_remainder - i
if row+1 >= n_row:
if col+1 >= n_col:
if next_total >= cutoff:
significance[0] += numpy.exp( const_part-next_total )
if significance[0] > significance_cutoff:
raise Cutoff_exceeded
else:
generate(0, col+1, col_sum[col+1], next_total)
else:
generate(row+1, col, col_remainder-i, next_total)
row_remainders[row] = old_row_remainder
try:
generate(0,0,col_sum[0],0.0)
except Cutoff_exceeded:
return None
return significance[0]
