def __call__(self, seq, result=None):
if result is None and self.working is None:
self.setResultArray(len(seq))
elif self.working is not None:
if len(seq) != self.working.shape[0]:
self.setResultArray(len(seq))
result = self.working
result.fill(0)
if type(seq) != str:
seq = ''.join(seq)
return seq_to_symbols(seq, self.motifs, self.motif_length, result)
def __call__(self, seq, result=None):
if result is None and self.working is None:
self.setResultArray(len(seq))
elif self.working is not None:
if len(seq) != self.working.shape[0]:
self.setResultArray(len(seq))
result = self.working
result.fill(0)
if type(seq) != str:
seq = ''.join(seq)
return seq_to_symbols(seq, self.motifs, self.motif_length, result)
def blockwise_bootstrap(signal, calc, block_size, num_reps, seq_to_symbols=None, num_stats=None):
"""returns observed statistic and the probability from the bootstrap
test of observing more `power' by chance than that estimated from the
observed signal
Arguments:
- signal: a series, can be a sequence object
- calc: function to calculate the period power, e.g. ipdft, hybrid,
auto_corr or any other statistic.
- block_size: size of contiguous values for resampling
- num_reps: number of randomly generated permutations
- seq_to_symbols: function to convert a sequence to 1/0. If not
provided, the raw data is used.
- num_stats: the number of statistics being evaluated for each
interation. Default to 1.
"""
signal_length = len(signal)
if seq_to_symbols is not None:
dtype='c'
else:
dtype=None # let numpy guess
signal = numpy.array(list(signal), dtype=dtype)
if seq_to_symbols is not None:
symbolic = seq_to_symbols(signal)
data = symbolic
else:
data = signal
obs_stat = calc(data)
if seq_to_symbols is not None:
if sum(symbolic) == 0:
p = [numpy.array([1.0, 1.0, 1.0]), 1.0][num_stats == 1]
return obs_stat, p
if num_stats is None:
try:
num_stats = calc.getNumStats()
except AttributeError:
num_stats = 1
if num_stats == 1:
count = 0
else:
count = numpy.zeros(num_stats)
for rep in range(num_reps):
# get sample positions
sampled_indices = sampled_places(block_size, signal_length)
new_signal = signal.take(sampled_indices)
if seq_to_symbols is not None:
symbolic = seq_to_symbols(new_signal)
data = symbolic
else:
data = new_signal
sim_stat = calc(data)
# count if > than observed
if num_stats > 1:
count[sim_stat >= obs_stat] += 1
elif sim_stat >= obs_stat:
count += 1
return obs_stat, count / num_reps
def blockwise_bootstrap(signal,
calc,
block_size,
num_reps,
seq_to_symbols=None,
num_stats=None):
"""returns observed statistic and the probability from the bootstrap
test of observing more `power' by chance than that estimated from the
observed signal
Arguments:
- signal: a series, can be a sequence object
- calc: function to calculate the period power, e.g. ipdft, hybrid,
auto_corr or any other statistic.
- block_size: size of contiguous values for resampling
- num_reps: number of randomly generated permutations
- seq_to_symbols: function to convert a sequence to 1/0. If not
provided, the raw data is used.
- num_stats: the number of statistics being evaluated for each
interation. Default to 1.
"""
signal_length = len(signal)
if seq_to_symbols is not None:
dtype = 'c'
else:
dtype = None # let numpy guess
signal = numpy.array(list(signal), dtype=dtype)
if seq_to_symbols is not None:
symbolic = seq_to_symbols(signal)
data = symbolic
else:
data = signal
obs_stat = calc(data)
if seq_to_symbols is not None:
if sum(symbolic) == 0:
p = [numpy.array([1.0, 1.0, 1.0]), 1.0][num_stats == 1]
return obs_stat, p
if num_stats is None:
try:
num_stats = calc.getNumStats()
except AttributeError:
num_stats = 1
if num_stats == 1:
count = 0
else:
count = numpy.zeros(num_stats)
for rep in range(num_reps):
# get sample positions
sampled_indices = sampled_places(block_size, signal_length)
new_signal = signal.take(sampled_indices)
if seq_to_symbols is not None:
symbolic = seq_to_symbols(new_signal)
data = symbolic
else:
data = new_signal
sim_stat = calc(data)
# count if > than observed
if num_stats > 1:
count[sim_stat >= obs_stat] += 1
elif sim_stat >= obs_stat:
count += 1
return obs_stat, count / num_reps