def bootstrap_cov(self, B, alpha=0.05, keep_seqids=None, progress_bar=False,
average_replicates=True, return_straps=False, ci_method='pivot',
**kwargs):
"""
Bootstrap whole covaraince matrix.
Params:
- B: number of bootstraps
- alpha: α level
- keep_seqids: which seqids to include in bootstrap; if None, all are used.
- return_straps: whether to return the actual bootstrap vectors.
- ci_method: 'pivot' or 'percentile'
- **kwargs: passed to calc_cov() and calc_cov_by_tile()
"""
# our statistic:
cov = self.calc_cov(keep_seqids=keep_seqids, **kwargs)
if average_replicates:
cov = np.nanmean(stack_temporal_covariances(cov, self.R, self.T), axis=2)
covs, het_denoms = self.calc_cov_by_tile(return_ratio_parts=True,
keep_seqids=keep_seqids, **kwargs)
covs, het_denoms = np.stack(covs), np.stack(het_denoms)
return block_bootstrap_ratio_averages(covs, het_denoms,
block_indices=self.tile_indices,
block_seqids=self.tile_df['seqid'].values,
estimator=cov_estimator,
statistic=cov,
B=B, alpha=alpha,
progress_bar=progress_bar,
keep_seqids=keep_seqids, return_straps=return_straps,
ci_method=ci_method,
# kwargs passed directly to cov_estimator
average_replicates=average_replicates,
R=self.R, T=self.T)
def cov_estimator(cov, het_denom, R, T, average_replicates=False, warn=False):
warn_type = 'ignore' if not warn else 'warn'
with np.errstate(divide=warn_type, invalid=warn_type):
cov = cov / het_denom
if not average_replicates:
return cov
else:
return np.mean(stack_temporal_covariances(cov, R, T), axis=2)
def calc_G(self, average_replicates=False, abs=False,
product_only=False,
use_masked=False):
covs = self.calc_cov(standardize=False,
product_only=product_only,
use_masked=use_masked)
# calculate the total variances for different ts
vars = []
for t in np.arange(1, self.T+1):
vars.append(np.stack(self.calc_var(t=t, standardize=False)))
total_vars = np.stack(vars, axis=0)
temp_covs = stack_temporal_covariances(covs, self.R, self.T)
G = G_estimator(temp_covs, total_vars,
average_replicates=average_replicates, abs=abs)
return G
def convergence_corr(covs, R, T):
replicate_covs = stack_replicate_covariances(covs, R, T, upper_only=False)
temporal_covs = stack_temporal_covariances(covs, R, T)
num, denom = (convergence_corr_numerator(replicate_covs[np.newaxis, ...]),
convergence_corr_denominator(temporal_covs[np.newaxis, ...]))
return num / denom