{{ document_title }}
{% if downloads %}{% for file in downloads %}{{ file.label }}{% if not loop.last %} · {% endif %}{% endfor %}{% endif %}{% extends "report_base.html" %} {% block title %}{{ title }}{% endblock %} {% block items %}mappings{% endblock %} {% block heading %}{{ fit_id }}{% endblock %} {% block subtitle %}{{ opid }}{% endblock %} {% block placeholder %}Search parameters, mappings, metrics …{% endblock %} {% block badges %}{% for badge in badges %}{{ badge.label }} {{ badge.value }}{% endfor %}{% endblock %} {# Two tabs: "Report" is the report document (methods, results, references) followed by the overview, the results and the identifiability of the fit, "Fit mappings" the figures of the cost and weight and of the residuals of every fit mapping, the table of the metrics of every fit mapping and a card per fit mapping. A tab is a link with `data-tab`, its panel a `.tab-panel` with `data-panel`; the page opens on the report, or on the tab of the element its address names. #} {% block head %} {% endblock %} {% block nav %} Report Fit mappings {% endblock %} {% block chips %} {% for kind in kinds %} {% endfor %} {% endblock %} {% block footer %}{% for file in files %}{{ file.label }}{% if not loop.last %} · {% endif %}{% endfor %}{% endblock %} {% block content %}
The PDF could not be created: {{ pdf_error }}. The Typst source report.typ is kept, every other output was written.
{{ document_title }}
{% if downloads %}{% for file in downloads %}{{ file.label }}{% if not loop.last %} · {% endif %}{% endfor %}{% endif %}| parameter | {% if versioned_parameters %}target | {% endif %} {% for pset in parameter_set_ids %}{{ pset }} | {% endfor %}lower | upper | unit |
|---|---|---|---|---|---|
| {{ row.pid }} | {% if versioned_parameters %}{{ row.target }} | {% endif %} {% for value in row.set_values %}{{ value }} | {% endfor %}{{ row.lower }} | {{ row.upper }} | {{ row.unit }} |
{% for message in bound_warnings %}{{ message }}
{% endfor %}
| network | pattern | layers | targets |
|---|---|---|---|
| {{ hook.name }} | {{ hook.kind }} | {{ hook.description }} | {{ hook.targets }} |
| array | elements? | estimated? | {% for pset in parameter_set_ids %}{{ pset }} min | {{ pset }} max | {{ pset }} norm? | {% endfor %}lower | upper |
|---|---|---|---|---|---|---|---|
| {{ row.label }} | {{ row.elements }} | {{ row.estimated }} | {% for values in row.set_values %}{% for value in values %}{{ value }} | {% endfor %}{% endfor %}{{ row.lower }} | {{ row.upper }} |
{% for message in array_bound_warnings %}{{ message }}
{% endfor %}
| setting | value |
|---|---|
| {{ key }} | {{ value }} |
| experiment | {% for kind in kinds %}{{ kind }} | {% endfor %}mappings |
|---|---|---|
| {{ row.experiment }} | {% for count in row.counts %}{{ count }} | {% endfor %}{{ row.total }} |
| total | {% for count in data_total.counts %}{{ count }} | {% endfor %}{{ data_total.total }} |
| {{ column.name }}{% if column.hint %}?{% endif %} | {% endfor %}
|---|
| {{ value }} | {% endfor %}
| {{ column }} | {% endfor %}
|---|
| {{ value }} | {% endfor %}
The curvature of the cost at the parameters, FIM = J'J with the jacobian of the weighted residuals, from which the standard errors and the correlations of the parameters follow. It is a local statement: the intervals are exact for a cost which is a quadratic around the optimum and approximate for a non-linear model, where the profile likelihood is the one to trust. The errors and the intervals are computed in the space the optimizer searches, so an interval on a logarithmic scale is not symmetric around the value.
The information does not have full rank: the data does not constrain every direction in parameter space, so the errors and the correlations of the unconstrained directions cannot be read. A parameter which no data informs is such a direction.
{% endif %}| {{ column.name }}{% if column.hint %}?{% endif %} | {% endfor %}
|---|
| {{ value }} | {% endfor %}
{{ fisher.note }}
{% endif %}| {% for pid in fisher.pids %} | {{ pid }} | {% endfor %}
|---|---|
| {{ pid }} | {% for value in fisher.correlation[loop.index0] %}{{ value }} | {% endfor %}
Profile likelihood of the parameters: every parameter is fixed at values around the optimum and the other parameters are optimized again. A parameter is identifiable if its profile crosses the threshold of the confidence level on both sides of the optimum, i.e., its confidence interval is finite; practically non-identifiable if the profile stays below the threshold up to a bound of the parameter; structurally non-identifiable if the profile is flat. An open side of an interval is shown as the bound of the parameter.
{% if identifiability.better_optimum %}A profile found a cost below the cost of the parameter set, the fit did not converge to the optimum. The threshold is relative to the lowest cost.
{% endif %}| parameter | value | ci lower | ci upper | unit | identifiability | points |
|---|---|---|---|---|---|---|
| {{ row.pid }} | {{ row.value }} | {{ row.ci_lower }} | {{ row.ci_upper }} | {{ row.unit }} | {{ row.label }}{% if not row.converged %} not converged{% endif %} | {{ row.n_points }} |
| {{ column }}{% if hints.get(column) %}?{% endif %} | {% endfor %}||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| {{ row.parameter_set }} | {{ row.experiment }} | {{ row.mapping }} | {{ row.kind }} | {{ row.n }} | {{ row.mse }} | {{ row.rmse }} | {{ row.nrmse }} | {{ row.rmse_w }} | {{ row.r2 }} | {{ row.within_2fold }} |