def rawAggregateExperimentResults()

in analysis/analyze.py [0:0]

• 31 lines of code
• 2 McCabe index (conditional complexity)

def rawAggregateExperimentResults(dat, by=['qid', 'query_idx', 'query', 'sketch_name']):
    by = cleanByClause(by)
    aggdat = dat.groupby(by, as_index=False).agg(
        bias=('error', np.mean),
        std=('error', np.std),
        rmse=('error', lambda x: np.linalg.norm(x)),
        q5=('error',lambda x: np.quantile(x, 0.05)),
        q25=('error',lambda x: np.quantile(x, 0.25)),
        q75=('error',lambda x: np.quantile(x, 0.75)),
        q95=('error',lambda x: np.quantile(x, 0.95)),
        size=('sketch_size', np.mean),
        size_bytes=('sketch_bytes', np.mean),
        observed_max_size=('sketch_size', np.max),
        count=('error', 'count'),
    )
        
    # two level calculation for variance
    vardat_by_workload = dat.groupby(by + ['data_seed'], as_index=False).agg(
        var=('error', populationVar),
        bias=('error', np.mean),
        count=('error','count'),
    )
    vardat_by_workload['var'] = vardat_by_workload['var'] / np.sqrt(vardat_by_workload['count']-1 + 1e-6)    
    
    vardat = vardat_by_workload.groupby(by, as_index=False).agg(
        var_expectation=('bias', populationVar),
        expected_var=('var', np.mean),
        count_by_seed=('bias', 'count')
    )
    vardat['var_bias'] = (vardat['var_expectation'] + vardat['expected_var']) / (vardat['count_by_seed']-1 + 1e-6)
    aggdat = aggdat.merge(vardat)
    
    aggdat['bias_lower95'] = aggdat['bias'] - 1.96 * np.sqrt(aggdat['var_bias']) 
    aggdat['bias_upper95'] = aggdat['bias'] + 1.96 * np.sqrt(aggdat['var_bias']) 
    
    return aggdat