localuf.data_processors

Module for functions to process numerical data, mainly from sim.

Functions

add_ignored_timesteps(data[, ...])	Add ignored timesteps to data.
get_failure_data(data[, p_slice, alpha, method])	Get failure stats from output of sim.accuracy.monte_carlo.
get_failure_data_from_subset_sample(data, ...)	Get failure stats from output of sim.accuracy.subset_sample.
get_log_runtime_data(data)	Get log runtime data from runtime data.
get_stats(log_data[, missing])	Get WLS (weighted least squares) stats from output of either get_log_runtime_data OR get_failure_data.

localuf.data_processors.get_failure_data(data, p_slice=slice(None, None, None), alpha=np.float64(0.31731050786291415), method='wilson')

Get failure stats from output of sim.accuracy.monte_carlo.

Parameters:

• data (DataFrame) – output from sim.accuracy.monte_carlo.

• p_slice – slice of probabilities to restrict output to.

• alpha – significance level of confidence intervals.

• method – method to compute confidence intervals. For details on confidence intervals, see https://www.statsmodels.org/dev/generated/statsmodels.stats.proportion.proportion_confint.html.

Return dT:

a DataFrame where each row a (distance, probability); columns are:

• f logical failure rate

• lo lower confidence bound of f

• hi upper confidence bound of f

• x log10(p)

• y log10(f)

• yerr half the confidence interval of y.

Return type:

DataFrame

localuf.data_processors.get_failure_data_from_subset_sample(data, code_class, noise, noise_levels, alpha=np.float64(0.31731050786291415), method='normal')

Get failure stats from output of sim.accuracy.subset_sample.

Returns:

A DataFrame indexed by (distance, noise level), with columns:

• f logical error probability.

• lo lower bound of f.

• hi upper bound of f.

Parameters:

• data (DataFrame)

• code_class (Type[Code])

• noise (Literal['code capacity', 'phenomenological', 'circuit-level'])

• noise_levels (ndarray[tuple[Any, ...], dtype[float64]])

• alpha (float)

Side effects: adds columns f, SE_lo, SE_hi to data.

localuf.data_processors.get_log_runtime_data(data)

Get log runtime data from runtime data.

data output from sim.runtime.batch.

Returns log_data:

A DataFrame where each column a stat: x, y, or yerr; row, a (probability, distance).

Parameters:

data (DataFrame)

localuf.data_processors.get_stats(log_data, missing='drop', **kwargs_for_WLS)

Get WLS (weighted least squares) stats from output of either get_log_runtime_data OR get_failure_data.

Parameters:

• log_data (DataFrame) – output from get_log_runtime_data or get_failure_data.

• missing – passed to statsmodels.regression.linear_model.WLS.

• kwargs_for_WLS – passed to statsmodels.regression.linear_model.WLS.

Return stats:

A DataFrame where each row an x-value (probability OR distance); columns are:

• intercept

• se_intercept

• gradient

• se_gradient

• r_squared

localuf.data_processors.add_ignored_timesteps(data, extra_steps_per_layer=2, layers_per_sample=<function <lambda>>)

Add ignored timesteps to data.

Parameters:

• data (DataFrame) – a DataFrame where each column a (distance, noise level); row, a runtime sample.

• extra_steps_per_layer – number of ignored timesteps per layer in the decoding graph.

• layers_per_sample (Callable[[int], int]) – a function with input d that outputs the measurement round count per sample.

E.g. if data is output of sim.runtime.frugal with time_only='merging' using Snowflake with the 1:1 schedule, the default kwargs of this function will convert it to the analogous output of sim.runtime.frugal with time_only='all'. This is because the extra timesteps per layer are a drop and a grow. If using Snowflake with the 2:1 schedule, set extra_steps_per_layer=3 as there is one additional grow timestep.