summary_metrics
Back to L1.5 | Browse all axes | Browse all options
Axis
summary_metricson sub-layerL1_5_B_univariate_summary(layerl1_5).
Sub-layer
L1_5_B_univariate_summary
Axis metadata
Default:
['mean', 'sd', 'min', 'max', 'n_missing']Sweepable: False
Status: operational
Operational status summary
Operational: 8 option(s)
Future: 0 option(s)
Options
kurtosis – operational
Sample excess kurtosis per series (fourth standardised moment, normal = 0).
Adds kurtosis to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Heavy-tail diagnostic; large values motivate winsorisation at L2.C or robust losses at L5.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, mean, sd, min, max, skew
Last reviewed 2026-05-05 by macroforecast author.
max – operational
Sample maximum per series.
Adds max to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Detecting outlier records prior to L2.C handling; suspicious upper bounds.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, mean, sd, min, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
mean – operational
Sample mean per series.
Adds mean to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
First-moment summary for level series; cross-series comparison of central tendency.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, sd, min, max, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
min – operational
Sample minimum per series.
Adds min to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Detecting clipping artifacts (e.g. a 0 sentinel) or suspicious lower bounds.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, mean, sd, max, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
n_missing – operational
Count of NaN entries per series.
Adds n_missing to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Quantifying imputation load before L2.D runs; high counts may justify dropping the series.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, mean, sd, min, max, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
n_obs – operational
Number of non-NaN observations per series.
Adds n_obs to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Pair with n_missing to spot heavily-missing predictors that L2.D will need to impute.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_missing, mean, sd, min, max, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
sd – operational
Sample standard deviation per series.
Adds sd to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Second-moment scale; informs whether L3 scale standardisation is necessary.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, mean, min, max, skew, kurtosis
Last reviewed 2026-05-05 by macroforecast author.
skew – operational
Sample skewness per series (third standardised moment).
Adds skew to the per-series summary table emitted by L1.5.B. summary_metrics is a multi-select axis – listing several metrics produces a wide-form table with one row per series and one column per chosen metric.
When to use
Identifying asymmetric distributions that may justify a log transform at L2.B.
References
macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’
Tukey (1977) ‘Exploratory Data Analysis’, Addison-Wesley.
Related options: n_obs, n_missing, mean, sd, min, max, kurtosis
Last reviewed 2026-05-05 by macroforecast author.