# Postprocessing and reporting¶

The ix modeling platform provides powerful features to perform calculations and other postprocessing after a message_ix.Scenario has been solved by the associated model. The MESSAGEix framework uses these features to provide zero-configuration reporting of models built on the framework.

These features are accessible through Reporter, which can produce multiple reports from one or more Scenarios. A report is identified by a key (usually a string), and may…

• perform arbitrarily complex calculations while intelligently handling units;

• read and make use of data that is ‘exogenous’ to (not included in) a Scenario;

• produce output as Python or R objects (in code), or to files or databases;

• calculate only a requested subset of quantities; and

• much, much more!

Contents:

## Terminology¶

ixmp.reporting handles numerical quantities, which are scalar (0-dimensional) or array (1 or more dimensions) data with optional associated units. ixmp parameters, scalars, equations, and time-series data all become quantities for the purpose of reporting.

Every quantity and report is identified by a key, which is a str or other hashable object. Special keys are used for multidimensional quantities. For instance: the MESSAGEix parameter resource_cost, defined with the dimensions (node n, commodity c, grade g, year y) is identified by the key 'resource_cost:n-c-g-y'. When summed across the grade/g dimension, it has dimensions n, c, y and is identified by the key 'resource_cost:n-c-y'.

Non-model 1 quantities and reports are produced by computations, which are atomic tasks that build on other computations. The most basic computations—for instance, resource_cost:n-c-g-y—simply retrieve raw/unprocessed data from a message_ix.Scenario and return it as a Quantity. Advanced computations can depend on many quantities, and/or combine quantities together into a structure like a document or spreadsheet. Computations are defined in ixmp.reporting.computations and message_ix.reporting.computations, but most common computations can be added using the methods of Reporter.

1

i.e. quantities that do not exist within the mathematical formulation of the model itself, and do not affect its solution.

## Basic usage¶

A reporting workflow has the following steps:

1. Obtain a message_ix.Scenario object from an ixmp.Platform.

2. Use from_scenario() to create a Reporter object.

3. (optionally) Use Reporter built-in methods or advanced features to add computations to the reporter.

4. Use get() to retrieve the results (or trigger the effects) of one or more computations.

>>> from ixmp import Platform
>>> from message_ix import Scenario, Reporter
>>>
>>> mp = Platform()
>>> scen = Scenario(scen)
>>> rep = Reporter.from_scenario(scen)
>>> rep.get('all')


Note

Reporter stores defined computations, but these are not executed until get() is called—or the results of one computation are required by another. This allows the Reporter to skip unneeded (and potentially slow) computations. A Reporter may contain computations for thousands of model quantities and derived quantities, but a call to get() may only execute a few of these.

## Customization¶

A Reporter prepared with from_scenario() always contains a key scenario, referring to the Scenario to be reported.

The method Reporter.add() can be used to add arbitrary Python code that operates directly on the Scenario object:

>>> def my_custom_report(scenario):
>>>     """Function with custom code that manipulates the *scenario*."""
>>>     print('foo')
>>>
>>> rep.get('custom')
foo


In this example, the function my_custom_report() could run to thousands of lines; read to and write from multiple files; invoke other programs or Python scripts; etc.

In order to take advantage of the performance-optimizing features of the Reporter, however, such calculations can be instead composed from atomic (i.e. small, indivisible) computations.

## Reporter, Key, and Quantity classes¶

 ixmp.reporting.Reporter(**kwargs) Class for generating reports on ixmp.Scenario objects. ixmp.reporting.Key(name[, dims, tag]) A hashable key for a quantity that includes its dimensionality. ixmp.reporting.Quantity Convert arguments to the internal Quantity data format.

The ixmp.Reporter automatically adds keys based on the contents of the ixmp.Scenario argument. The message_ix.reporting.Reporter adds additional keys for derived quantities specific to the MESSAGEix model framework. These include:

• out: the product of output (output efficiency) and ACT (activity).

• out_hist = output × ref_activity (historical reference activity),

• in = input × ACT,

• in_hist = input × ref_activity,

• emi = emission_factor × ACT,

• emi_hist = emission_factor × ref_activity,

• inv = inv_cost × CAP_NEW,

• inv_hist = inv_cost × ref_new_capacity,

• fom = fix_cost × CAP,

• fom_hist = fix_cost × ref_capacity,

• vom = var_cost × ACT, and

• vom_hist = var_cost × ref_activity.

• tom = fom + vom.

• land_out = land_output × LAND,

• land_use_qty = land_use × LAND,

• land_emi = land_emission × LAND,

• addon conversion, the model parameter addon_conversion (note space versus underscore), except broadcast across individual add-on technologies (ta) rather than add-on types (type_addon),

• addon up, which is addon_up similarly broadcast.,

• addon ACT = addon conversion × ACT,

• addon in = input × addon ACT,

• addon out = output × addon ACT, and

• addon potential = addon up × addon ACT, the maximum potential activity by add-on technology.

• price emission, the model variable PRICE_EMISSION broadcast across emission species (e) and technologies (t) rather than types (type_emission, type_tec).

Tip

Use full_key() to retrieve the full-dimensionality Key for any of these quantities.

• <name>:pyam for the above quantities, plus:

• CAP:pyam (from CAP)

• CAP_NEW:pyam (from CAP_NEW)

These keys return the values in the IAMC data format, as pyam objects.

• map_<name> as ‘indicator’ quantities for the mapping sets cat_<name>.

• Standard reports message:system, message_costs, and message:emissions.

• The report message:default, collecting all of the above reports.

These automatic features of Reporter are controlled by:

class ixmp.reporting.Reporter(**kwargs)

Class for generating reports on ixmp.Scenario objects.

A Reporter is used to postprocess data from from one or more ixmp.Scenario objects. The get() method can be used to:

• Retrieve individual quantities. A quantity has zero or more dimensions and optional units. Quantities include the ‘parameters’, ‘variables’, ‘equations’, and ‘scalars’ available in an ixmp.Scenario.

• Generate an entire report composed of multiple quantities. A report may:

• Read in non-model or exogenous data,

• Trigger output to files(s) or a database, or

• Execute user-defined methods.

Every report and quantity (including the results of intermediate steps) is identified by a utils.Key; all the keys in a Reporter can be listed with keys().

Reporter uses a graph data structure to keep track of computations, the atomic steps in postprocessing: for example, a single calculation that multiplies two quantities to create a third. The graph allows get() to perform only the requested computations. Advanced users may manipulate the graph directly; but common reporting tasks can be handled by using Reporter methods:

 add(data, *args, **kwargs) General-purpose method to add computations. add_file(path[, key]) Add exogenous quantities from path. add_product(key, *quantities[, sums]) Add a computation that takes the product of quantities. aggregate(qty, tag, dims_or_groups[, …]) Add a computation that aggregates qty. apply(generator, *keys, **kwargs) Add computations by applying generator to keys. check_keys(*keys) Check that keys are in the Reporter. configure([path]) Configure the Reporter. describe([key, quiet]) Return a string describing the computations that produce key. disaggregate(qty, new_dim[, method, args]) Add a computation that disaggregates qty using method. finalize(scenario) Prepare the Reporter to act on scenario. full_key(name_or_key) Return the full-dimensionality key for name_or_key. get([key]) Execute and return the result of the computation key. Return the keys of graph. set_filters(**filters) Apply filters ex ante (before computations occur). visualize(filename, **kwargs) Generate an image describing the reporting structure. write(key, path) Write the report key to the file path.
graph = {'config': {}}

add(data, *args, **kwargs)

add() can be called in several ways; its behaviour depends on data; see below. It chains to methods such as add_single(), add_queue(), and apply(), which can also be called directly.

Parameters
• data (various) –

• args (various) –

Other Parameters

sums (bool, optional) – If True, all partial sums of the key data are also added to the Reporter.

Returns

Some or all of the keys added to the Reporter.

Return type

list of Key-like

Raises

KeyError – If a target key is already in the Reporter; any key referred to by a computation does not exist; or sums=True and the key for one of the partial sums of key is already in the Reporter.

add() may be used to:

• Provide an alias from one key to another:

>>> r.add('aliased name', 'original name')

• Define an arbitrarily complex computation in a Python function that operates directly on the ixmp.Scenario:

>>> def my_report(scenario):
>>>     # many lines of code
>>>     return 'foo'
>>> r.finalize(scenario)
>>> r.get('my report')
foo


Note

Use care when adding literal str values (2); these may conflict with keys that identify the results of other computations.

add_file(path, key=None, **kwargs)

Reporting the key or using it in other computations causes path to be loaded and converted to Quantity.

Parameters
• path (os.PathLike) – Path to the file, e.g. ‘/path/to/foo.ext’.

• key (str or Key, optional) – Key for the quantity read from the file.

Other Parameters
• dims (dict or list or set) – Either a collection of names for dimensions of the quantity, or a mapping from names appearing in the input to dimensions.

• units (str or pint.Unit) – Units to apply to the loaded Quantity.

Returns

Either key (if given) or e.g. file:foo.ext based on the path name, without directory components.

Return type

Key

add_load_file(path, key=None, **kwargs)

Reporting the key or using it in other computations causes path to be loaded and converted to Quantity.

Parameters
• path (os.PathLike) – Path to the file, e.g. ‘/path/to/foo.ext’.

• key (str or Key, optional) – Key for the quantity read from the file.

Other Parameters
• dims (dict or list or set) – Either a collection of names for dimensions of the quantity, or a mapping from names appearing in the input to dimensions.

• units (str or pint.Unit) – Units to apply to the loaded Quantity.

Returns

Either key (if given) or e.g. file:foo.ext based on the path name, without directory components.

Return type

Key

add_product(key, *quantities, sums=True)

Add a computation that takes the product of quantities.

Parameters
• key (str or Key) – Key of the new quantity. If a Key, any dimensions are ignored; the dimensions of the product are the union of the dimensions of quantities.

• sums (bool, optional) – If True, all partial sums of the new quantity are also added.

Returns

The full key of the new quantity.

Return type

Key

add_queue(queue, max_tries=1, fail='raise')

Parameters
• queue (list of 2-tuple) – The members of each tuple are the arguments (i.e. a list or tuple) and keyword arguments (i.e. a dict) to add().

• max_tries (int, optional) – Retry adding elements up to this many times.

• fail ('raise' or log level, optional) – Action to take when a computation from queue cannot be added after max_tries.

add_single(key, *computation, strict=False, index=False)

Add a single computation at key.

Parameters
• key (str or Key or hashable) – A string, Key, or other value identifying the output of task.

• computation (object) –

Any dask computation, i.e. one of:

1. any existing key in the Reporter.

2. any other literal value or constant.

3. a task, i.e. a tuple with a callable followed by one or more computations.

4. A list containing one or more of #1, #2, and/or #3.

• strict (bool, optional) – If True, key must not already exist in the Reporter, and any keys referred to by computation must exist.

• index (bool, optional) – If True, key is added to the index as a full-resolution key, so it can be later retrieved with full_key().

aggregate(qty, tag, dims_or_groups, weights=None, keep=True, sums=False)

Add a computation that aggregates qty.

Parameters
• qty (Key or str) – Key of the quantity to be aggregated.

• tag (str) – Additional string to add to the end the key for the aggregated quantity.

• dims_or_groups (str or iterable of str or dict) – Name(s) of the dimension(s) to sum over, or nested dict.

• weights (xarray.DataArray, optional) – Weights for weighted aggregation.

• keep (bool, optional) – Passed to computations.aggregate.

• sums (bool, optional) – Passed to add().

Returns

The key of the newly-added node.

Return type

Key

apply(generator, *keys, **kwargs)

Add computations by applying generator to keys.

Parameters
• generator (callable) – Function to apply to keys.

• keys (hashable) – The starting key(s).

• kwargs – Keyword arguments to generator.

check_keys(*keys)

Check that keys are in the Reporter.

If any of keys is not in the Reporter, KeyError is raised. Otherwise, a list is returned with either the key from keys, or the corresponding full_key().

configure(path=None, **config)

Configure the Reporter.

Accepts a path to a configuration file and/or keyword arguments. Configuration keys loaded from file are replaced by keyword arguments.

Valid configuration keys include:

Warns

UserWarning – If config contains unrecognized keys.

default_key = None

The default reporting key.

describe(key=None, quiet=True)

Return a string describing the computations that produce key.

If key is not provided, all keys in the Reporter are described.

The string can be printed to the console, if not quiet.

disaggregate(qty, new_dim, method='shares', args=[])

Add a computation that disaggregates qty using method.

Parameters
• qty (hashable) – Key of the quantity to be disaggregated.

• new_dim (str) – Name of the new dimension of the disaggregated variable.

• method (callable or str) – Disaggregation method. If a callable, then it is applied to var with any extra args. If then a method named ‘disaggregate_{method}’ is used.

• args (list, optional) – Additional arguments to the method. The first element should be the key for a quantity giving shares for disaggregation.

Returns

The key of the newly-added node.

Return type

Key

finalize(scenario)

Prepare the Reporter to act on scenario.

The Scenario object scenario is associated with the key 'scenario'. All subsequent processing will act on data from this scenario.

classmethod from_scenario(scenario, **kwargs)

Create a Reporter by introspecting scenario.

Parameters
• scenario (ixmp.Scenario) – Scenario to introspect in creating the Reporter.

• kwargs (optional) – Passed to Scenario.configure().

Returns

A Reporter instance containing:

• A ‘scenario’ key referring to the scenario object.

• Each parameter, equation, and variable in the scenario.

• All possible aggregations across different sets of dimensions.

• Each set in the scenario.

Return type

Reporter

full_key(name_or_key)

Return the full-dimensionality key for name_or_key.

An ixmp variable ‘foo’ with dimensions (a, c, n, q, x) is available in the Reporter as 'foo:a-c-n-q-x'. This Key can be retrieved with:

rep.full_key('foo')
rep.full_key('foo:c')
# etc.

get(key=None)

Execute and return the result of the computation key.

Only key and its dependencies are computed.

Parameters

key (str, optional) – If not provided, default_key is used.

Raises

ValueError – If key and default_key are both None.

keys()

Return the keys of graph.

set_filters(**filters)

Apply filters ex ante (before computations occur).

Filters are stored in the reporter at the 'filters' key, and are passed to ixmp.Scenario.par() and similar methods. All quantity values read from the Scenario are filtered before any other computations take place.

Parameters

filters (mapping of str → (list of str or None)) –

Argument names are dimension names; values are lists of allowable labels along the respective dimension, or None to clear any existing filters for the dimension.

If no arguments are provided, all filters are cleared.

property unit_registry

The pint.UnitRegistry() used by the Reporter.

visualize(filename, **kwargs)

Generate an image describing the reporting structure.

This is a shorthand for dask.visualize(). Requires graphviz.

write(key, path)

Write the report key to the file path.

class ixmp.reporting.Key(name, dims=[], tag=None)

A hashable key for a quantity that includes its dimensionality.

Quantities in a Scenario can be indexed by one or more dimensions. Keys refer to quantities, using three components:

1. a string name,

2. zero or more ordered dimensions dims, and

3. an optional tag.

For example, an ixmp parameter with three dimensions can be initialized with:

>>> scenario.init_par('foo', ['a', 'b', 'c'], ['apple', 'bird', 'car'])


Key allows a specific, explicit reference to various forms of “foo”:

• in its full resolution, i.e. indexed by a, b, and c:

>>> k1 = Key('foo', ['a', 'b', 'c'])
>>> k1
<foo:a-b-c>

• in a partial sum over one dimension, e.g. summed across dimension c, with remaining dimensions a and b:

>>> k2 = k1.drop('c')
>>> k2
<foo:a-b>

• in a partial sum over multiple dimensions, etc.:

>>> k1.drop('a', 'c') == k2.drop('a') == 'foo:b'
True

• after it has been manipulated by different reporting computations, e.g.

>>> k3 = k1.add_tag('normalized')
>>> k3
<foo:a-b-c:normalized>
>>> k4
<foo:a-b-c:normalized+rescaled>


Notes:

A Key has the same hash, and compares equal to its str representation. repr(key) prints the Key in angle brackets (‘<>’) to signify that it is a Key object.

>>> str(k1)
'foo:a-b-c'
>>> repr(k1)
'<foo:a-b-c>'
>>> hash(k1) == hash('foo:a-b-c')
True


Keys are immutable: the properties name, dims, and tag are read-only, and the methods append(), drop(), and add_tag() return new Key objects.

Keys may be generated concisely by defining a convenience method:

>>> def foo(dims):
>>>     return Key('foo', dims.split())
>>> foo('a b c')
<foo:a-b-c>

add_tag(tag)

Return a new Key with tag appended.

append(*dims)

Return a new Key with additional dimensions dims.

property dims

Dimensions of the quantity, tuple of str.

drop(*dims)

Return a new Key with dims dropped.

classmethod from_str_or_key(value, drop=[], append=[], tag=None)

Return a new Key from value.

Parameters
Returns

Return type

Key

iter_sums()

Generate (key, task) for all possible partial sums of the Key.

property name

Name of the quantity, str.

classmethod product(new_name, *keys, tag=None)

Return a new Key that has the union of dimensions on keys.

Dimensions are ordered by their first appearance:

1. First, the dimensions of the first of the keys.

2. Next, any additional dimensions in the second of the keys that were not already added in step 1.

3. etc.

Parameters

new_name (str) – Name for the new Key. The names of keys are discarded.

property tag

Quantity tag, str.

ixmp.reporting.Quantity(data, *args, **kwargs)

Convert arguments to the internal Quantity data format.

Parameters
• data – Quantity data.

• args – Positional arguments, passed to AttrSeries or SparseDataArray.

• kwargs – Keyword arguments, passed to AttrSeries or SparseDataArray.

Other Parameters
• name (str, optional) – Quantity name.

• units (str, optional) – Quantity units.

• attrs (dict, optional) – Dictionary of attributes; similar to attrs.

The Quantity constructor converts its arguments to an internal, xarray.DataArray-like data format:

# Existing data
data = pd.Series(...)

# Convert to a Quantity for use in reporting calculations
qty = Quantity(data, name="Quantity name", units="kg")


## Computations¶

### Inherited from ixmp¶

Elementary computations for reporting.

Unless otherwise specified, these methods accept and return Quantity objects for data arguments/return values.

Calculations:

 add(*quantities[, fill_value]) Sum across multiple quantities. aggregate(quantity, groups, keep) Aggregate quantity by groups. apply_units(qty, units[, quiet]) Simply apply units to qty. disaggregate_shares(quantity, shares) Disaggregate quantity by shares. product(*quantities) Return the product of any number of quantities. ratio(numerator, denominator) Return the ratio numerator / denominator. select(qty, indexers[, inverse]) Select from qty based on indexers. sum(quantity[, weights, dimensions]) Sum quantity over dimensions, with optional weights.

Input and output:

 load_file(path[, dims, units]) Read the file at path and return its contents as a Quantity. write_report(quantity, path) Write a quantity to a file.

Data manipulation:

 concat(*objs, **kwargs) Concatenate Quantity objs.
ixmp.reporting.computations.aggregate(quantity, groups, keep)

Aggregate quantity by groups.

Parameters
• quantity (Quantity) –

• groups (dict of dict) – Top-level keys are the names of dimensions in quantity. Second-level keys are group names; second-level values are lists of labels along the dimension to sum into a group.

• keep (bool) – If True, the members that are aggregated into a group are returned with the group sums. If False, they are discarded.

Returns

Same dimensionality as quantity.

Return type

Quantity

ixmp.reporting.computations.apply_units(qty, units, quiet=False)

Simply apply units to qty.

Logs on level WARNING if qty already has existing units.

Parameters
ixmp.reporting.computations.concat(*objs, **kwargs)

Concatenate Quantity objs.

Any strings included amongst args are discarded, with a logged warning; these usually indicate that a quantity is referenced which is not in the Reporter.

ixmp.reporting.computations.data_for_quantity(ix_type, name, column, scenario, config)

Retrieve data from scenario.

Parameters
• ix_type ('equ' or 'par' or 'var') – Type of the ixmp object.

• name (str) – Name of the ixmp object.

• column ('mrg' or 'lvl' or 'value') – Data to retrieve. ‘mrg’ and ‘lvl’ are valid only for ix_type='equ', and ‘level’ otherwise.

• scenario (ixmp.Scenario) – Scenario containing data to be retrieved.

• config (dict of (str -> dict)) – The key ‘filters’ may contain a mapping from dimensions to iterables of allowed values along each dimension. The key ‘units’/’apply’ may contain units to apply to the quantity; any such units overwrite existing units, without conversion.

Returns

Data for name.

Return type

Quantity

ixmp.reporting.computations.disaggregate_shares(quantity, shares)

Disaggregate quantity by shares.

ixmp.reporting.computations.load_file(path, dims={}, units=None)

Read the file at path and return its contents as a Quantity.

Some file formats are automatically converted into objects for direct use in reporting code:

.csv:

Converted to Quantity. CSV files must have a ‘value’ column; all others are treated as indices, except as given by dims. Lines beginning with ‘#’ are ignored.

Parameters
• path (pathlib.Path) – Path to the file to read.

• dims (collections.abc.Collection or collections.abc.Mapping, optional) – If a collection of names, other columns besides these and ‘value’ are discarded. If a mapping, the keys are the column labels in path, and the values are the target dimension names.

• units (str or pint.Unit) – Units to apply to the loaded Quantity.

ixmp.reporting.computations.product(*quantities)

Return the product of any number of quantities.

ixmp.reporting.computations.ratio(numerator, denominator)

Return the ratio numerator / denominator.

Parameters
• numerator (Quantity) –

• denominator (Quantity) –

ixmp.reporting.computations.select(qty, indexers, inverse=False)

Select from qty based on indexers.

Parameters
• qty (Quantity) –

• indexers (dict (str -> list of str)) – Elements to be selected from qty. Mapping from dimension names to labels along each dimension.

• inverse (bool, optional) – If True, remove the items in indexers instead of keeping them.

ixmp.reporting.computations.sum(quantity, weights=None, dimensions=None)

Sum quantity over dimensions, with optional weights.

Parameters
• quantity (Quantity) –

• weights (Quantity, optional) – If dimensions is given, weights must have at least these dimensions. Otherwise, any dimensions are valid.

• dimensions (list of str, optional) – If not provided, sum over all dimensions. If provided, sum over these dimensions.

ixmp.reporting.computations.write_report(quantity, path)

Write a quantity to a file.

Parameters

path (str or Path) – Path to the file to be written.

## Configuration¶

 Configure reporting globally. ixmp.reporting.utils.RENAME_DIMS Dimensions to rename when extracting raw data from Scenario objects. ixmp.reporting.utils.REPLACE_UNITS Replacements to apply to quantity units before parsing by pint.
reporting.configure(**config)

Configure reporting globally.

Modifies global variables that affect the behaviour of all Reporters and computations, namely RENAME_DIMS and REPLACE_UNITS.

Valid configuration keys—passed as config keyword arguments—include:

Other Parameters
Warns

UserWarning – If config contains unrecognized keys.

ixmp.reporting.utils.RENAME_DIMS = {}

Dimensions to rename when extracting raw data from Scenario objects. Mapping from Scenario dimension name -> preferred dimension name. message_ix adds the standard short symbols for MESSAGE sets to this variable.

ixmp.reporting.utils.REPLACE_UNITS = {'%': 'percent'}

Replacements to apply to quantity units before parsing by pint. Mapping from original unit -> preferred unit.

## Utilities¶

ixmp.reporting.quantity.assert_quantity(*args)

Assert that each of args is a Quantity object.

Raises

TypeError – with a indicative message.

ixmp.reporting.utils.clean_units(input_string)

Tolerate messy strings for units.

Handles two specific cases found in MESSAGEix test cases:

• Dimensions enclosed in ‘[]’ have these characters stripped.

• The ‘%’ symbol cannot be supported by pint, because it is a Python operator; it is translated to ‘percent’.

ixmp.reporting.utils.collect_units(*args)

Return an list of ‘_unit’ attributes for args.

ixmp.reporting.utils.dims_for_qty(data)

Return the list of dimensions for data.

If data is a pandas.DataFrame, its columns are processed; otherwise it must be a list.

ixmp.reporting.RENAME_DIMS is used to rename dimensions.

ixmp.reporting.utils.filter_concat_args(args)

Filter out str and Key from args.

A warning is logged for each element removed.

ixmp.reporting.utils.parse_units(units_series)

Return a pint.Unit for a pd.Series of strings.