Skip to content

Latest commit

 

History

History
672 lines (521 loc) · 29.1 KB

decorators.md

File metadata and controls

672 lines (521 loc) · 29.1 KB
Raw

Decorators

While the 1:1 mapping of column -> function implementation is powerful, we've implemented a few decorators to promote business-logic reuse. The decorators we've defined are as follows (source can be found in function_modifiers):

@parameterize

Expands a single function into n, each of which correspond to a function in which the parameter value is replaced either by:

  1. A specified value
  2. The value from a specified upstream node.

Note that this can take the place of any of the @parameterize decorators below. In fact, they delegate to this!

import pandas as pd
from hamilton.function_modifiers import parameterize
from hamilton.function_modifiers import value, source


@parameterize(
    D_ELECTION_2016_shifted=dict(n_off_date=source('D_ELECTION_2016'), shift_by=value(3)),
    SOME_OUTPUT_NAME=dict(n_off_date=source('SOME_INPUT_NAME'), shift_by=value(1)),
)
def date_shifter(n_off_date: pd.Series, shift_by: int = 1) -> pd.Series:
    """{one_off_date} shifted by shift_by to create {output_name}"""
    return n_off_date.shift(shift_by)

By choosing literal or upstream, you can determine the source of your dependency. Note that you can also pass documentation. If you don't, it will use the parameterized docstring.

@parameterize(
    D_ELECTION_2016_shifted=(dict(n_off_date=source('D_ELECTION_2016'), shift_by=value(3)), "D_ELECTION_2016 shifted by 3"),
    SOME_OUTPUT_NAME=(dict(n_off_date=source('SOME_INPUT_NAME'), shift_by=value(1)),"SOME_INPUT_NAME shifted by 1")
)
def date_shifter(n_off_date: pd.Series, shift_by: int=1) -> pd.Series:
    """{one_off_date} shifted by shift_by to create {output_name}"""
    return n_off_date.shift(shift_by)

@parameterize_values (replacing @parametrized)

Expands a single function into n, each of which corresponds to a function in which the parameter value is replaced by that specific value.

import pandas as pd
from hamilton.function_modifiers import parameterize_values
import internal_package_with_logic

ONE_OFF_DATES = {
     #output name        # doc string               # input value to function
    ('D_ELECTION_2016', 'US Election 2016 Dummy'): '2016-11-12',
    ('SOME_OUTPUT_NAME', 'Doc string for this thing'): 'value to pass to function',
}
            # parameter matches the name of the argument in the function below
@parameterize_values(parameter='one_off_date', assigned_output=ONE_OFF_DATES)
def create_one_off_dates(date_index: pd.Series, one_off_date: str) -> pd.Series:
    """Given a date index, produces a series where a 1 is placed at the date index that would contain that event."""
    one_off_dates = internal_package_with_logic.get_business_week(one_off_date)
    return internal_package_with_logic.bool_to_int(date_index.isin([one_off_dates]))

We see here that parameterized allows you keep your code DRY by reusing the same function to create multiple distinct outputs. The parameter key word argument has to match one of the arguments in the function. The rest of the arguments are pulled from outside the DAG. The assigned_output key word argument takes in a dictionary of tuple(Output Name, Documentation string) -> value.

Note that @parametrized is deprecated, and we intend for you to use @parameterize_vales. We're consolidating to make the parameterization decorators more consistent! You have plenty of time to migrate, we wont make this a hard change until we have a Hamilton 2.0.0 to release.

@parameterize_sources (replacing @parameterized_inputs)

Expands a single function into n, each of which corresponds to a function in which the parameters specified are mapped to the specified inputs. Note this decorator and @parameterize_values are quite similar, except that the input here is another DAG node(s), i.e. column/input, rather than a specific scalar/static value.

import pandas as pd
from hamilton.function_modifiers import parameterize_sources


@parameterize_sources(
    D_ELECTION_2016_shifted=dict(one_off_date='D_ELECTION_2016'),
    SOME_OUTPUT_NAME=dict(one_off_date='SOME_INPUT_NAME')
)
def date_shifter(one_off_date: pd.Series) -> pd.Series:
    """{one_off_date} shifted by 1 to create {output_name}"""
    return one_off_date.shift(1)

We see here that parameterize_sources allows you to keep your code DRY by reusing the same function to create multiple distinct outputs. The key word arguments passed have to have the following structure:

OUTPUT_NAME = Mapping of function argument to input that should go into it.

So in the example, D_ELECTION_2016_shifted is an output that will correspond to replacing one_off_date with D_ELECTION_2016. Then similarly SOME_OUTPUT_NAME is an output that will correspond to replacing one_off_date with SOME_INPUT_NAME. The documentation for both uses the same function doc and will replace values that are templatized with the input parameter names, and the reserved value output_name.

To help visualize what the above is doing, it is equivalent to writing the following two function definitions:

def D_ELECTION_2016_shifted(D_ELECTION_2016: pd.Series) -> pd.Series:
    """D_ELECTION_2016 shifted by 1 to create D_ELECTION_2016_shifted"""
    return D_ELECTION_2016.shift(1)

def SOME_OUTPUT_NAME(SOME_INPUT_NAME: pd.Series) -> pd.Series:
    """SOME_INPUT_NAME shifted by 1 to create SOME_OUTPUT_NAME"""
    return SOME_INPUT_NAME.shift(1)

Note that @parameterized_inputs is deprecated, and we intend for you to use @parameterize_sources. We're consolidating to make the parameterization decorators more consistent! But we will not break your workflow for a long time.

Note: that the different input variables must all have compatible types with the original decorated input variable.

Migrating @parameterized*

As we've said above, we're planning on deprecating the following:

  • @parameterized_inputs (replaced by @parameterize_sources)
  • @parametrized (replaced by @parameterize_values, as that's what its really doing)
  • @parametrized_input (deprecated long ago, migrate to @parameterize_sources as that is more versatile.)

In other words, we're aligning around the following @parameterize implementations:

  • @parameterize -- this does everything you want
  • @parameterize_values -- this just changes the values, does not change the input source
  • @parameterize_sources-- this just changes the source of the inputs. We also changed the name from inputs -> sources as it was clearer (values are inputs as well).

The only non-drop-in change you'll have to do is for @parameterized. We won't update this until hamilton==2.0.0, though, so you'll have time to migrate for a while.

@extract_columns

This works on a function that outputs a dataframe, that we want to extract the columns from and make them individually available for consumption. So it expands a single function into n functions, each of which take in the output dataframe and output a specific column as named in the extract_columns decorator.

import pandas as pd
from hamilton.function_modifiers import extract_columns

@extract_columns('fiscal_date', 'fiscal_week_name', 'fiscal_month', 'fiscal_quarter', 'fiscal_year')
def fiscal_columns(date_index: pd.Series, fiscal_dates: pd.DataFrame) -> pd.DataFrame:
    """Extracts the fiscal column data.
    We want to ensure that it has the same spine as date_index.
    :param fiscal_dates: the input dataframe to extract.
    :return:
    """
    df = pd.DataFrame({'date_index': date_index}, index=date_index.index)
    merged = df.join(fiscal_dates, how='inner')
    return merged

Note: if you have a list of columns to extract, then when you call @extract_columns you should call it with an asterisk like this:

import pandas as pd
from hamilton.function_modifiers import extract_columns

@extract_columns(*my_list_of_column_names)
def my_func(...) -> pd.DataFrame:
   """..."""

@does

@does is a decorator that allows you to replace the decorated function with the behavior from another function. This allows for easy code-reuse when building repeated logic. You do this by decorating a function with@does, which takes in two parameters:

  1. replacing_function Required -- a function that takes in a "compatible" set of arguments. This means that it will work when passing the corresponding keyword arguments to the decorated function.
  2. **argument_mapping -- a mapping of arguments from the replacing function to the replacing function. This makes for easy reuse of functions. Confused? See the examples below.
import pandas as pd
from hamilton.function_modifiers import does

def _sum_series(**series: pd.Series) -> pd.Series:
    """This function takes any number of inputs and sums them all together."""
    return sum(series)

@does(_sum_series)
def D_XMAS_GC_WEIGHTED_BY_DAY(D_XMAS_GC_WEIGHTED_BY_DAY_1: pd.Series,
                              D_XMAS_GC_WEIGHTED_BY_DAY_2: pd.Series) -> pd.Series:
    """Adds D_XMAS_GC_WEIGHTED_BY_DAY_1 and D_XMAS_GC_WEIGHTED_BY_DAY_2"""
    pass

In the above example @does applies _sum_series to the function D_XMAS_GC_WEIGHTED_BY_DAY. Note we don't need any parameter replacement as _sum_series takes in just **kwargs, enabling it to work with any set of parameters (and thus any old function).

import pandas as pd
from hamilton.function_modifiers import does

import internal_company_logic

def _load_data(db: str, table: str) -> pd.DataFrame:
    """Helper function to load data using your internal company logic"""
    return internal_company_logic.read_table(db=db, table=table)

@does(_load_data, db='marketing_spend_db', table='marketing_spend_table')
def marketing_spend_data(marketing_spend_db: str, marketing_spend_table: str) -> pd.Series:
    """Loads marketing spend data from the database"""
    pass

@does(_load_data, db='client_acquisition_db', table='client_acquisition_table')
def client_acquisition_data(client_acquisition_db: str, client_acquisition_table: str) -> pd.Series:
    """Loads client acquisition data from the database"""
    pass

In the above example, @does applies our internal function _load_data, which applies custom logic to load a table from a database in the data warehouse. Note that we map the parameters -- in the first example, the value of the parameter marketing_spend_db is passed to db, and the value of the parameter marketing_spend_table is passed to table.

@config.when*

@config.when allows you to specify different implementations depending on configuration parameters.

The following use cases are supported:

  1. A column is present for only one value of a config parameter -- in this case, we define a function only once, with a @config.when
    import pandas as pd
    from hamilton.function_modifiers import config

    # signups_parent_before_launch is only present in the kids business line
    @config.when(business_line='kids')
    def signups_parent_before_launch(signups_from_existing_womens_tf: pd.Series) -> pd.Series:
        """TODO:
        :param signups_from_existing_womens_tf:
        :return:
        """
        return signups_from_existing_womens_tf
  1. A column is implemented differently for different business inputs, e.g. in the case of Stitch Fix gender intent.
    import pandas as pd
    from hamilton.function_modifiers import config, model
    import internal_package_with_logic

    # Some 21 day autoship cadence does not exist for kids, so we just return 0s
    @config.when(gender_intent='kids')
    def percent_clients_something__kids(date_index: pd.Series) -> pd.Series:
        return pd.Series(index=date_index.index, data=0.0)

    # In other business lines, we have a model for it
    @config.when_not(gender_intent='kids')
    @model(internal_package_with_logic.GLM, 'some_model_name', output_column='percent_clients_something')
    def percent_clients_something_model() -> pd.Series:
        pass

Note the following:

  • The function cannot have the same name in the same file (or python gets unhappy), so we name it with a __ (dunderscore) as a suffix. The dunderscore is removed before it goes into the DAG.

  • There is currently no @config.otherwise(...) decorator, so make sure to have config.when specify set of configuration possibilities. Any missing cases will not have that output column (and subsequent downstream nodes may error out if they ask for it). To make this easier, we have a few more @config decorators:

    • @config.when_not(param=value) Will be included if the parameter is not equal to the value specified.
    • @config.when_in(param=[value1, value2, ...]) Will be included if the parameter is equal to one of the specified values.
    • @config.when_not_in(param=[value1, value2, ...]) Will be included if the parameter is not equal to any of the specified values.
    • @config If you're feeling adventurous, you can pass in a lambda function that takes in the entire configuration and resolves to True or False. You probably don't want to do this.

To pass in the right value, you would provide param, e.g. gender_intent, or business_line, as a field in the dictionary passed to instantiate the driver. E.g.

config = {
  "business_line": "kids"
}
dr = driver.Driver(config, module1, ...)

@tag and friends

@tag

Allows you to attach metadata to an output(s), i.e. all nodes generated by a function and its decorators (note, this, by default, only applies to "final" nodes -- not any intermediate nodes that are generated...). See "A mental model for decorators" below for how/why this works.

A common use of this is to enable marking nodes as part of some data product, or for GDPR/privacy purposes.

For instance:

import pandas as pd
from hamilton.function_modifiers import tag

def intermediate_column() -> pd.Series:
    pass

@tag(data_product='final', pii='true')
def final_column(intermediate_column: pd.Series) -> pd.Series:
    pass

@tag also allows you to specify a target with the target_ parameter. See "A mental model for decorators" below for more details.

@tag_outputs (deprecated)

tag_outputs enables you to attach metadata to a function that outputs multiple nodes, and give different tag values to different outputs:

import pandas as pd
from hamilton.function_modifiers import tag_outputs, extract_columns

def intermediate_column() -> pd.Series:
    pass

@tag_outputs(
    public={'column_a' : 'public'},
    private={'column_b' : 'private'})
@extract_columns('column_a', 'column_b')
def data_used_in_multiple_ways() -> pd.DataFrame:
    return load_some_data(...)

In this case, the tag accessibility would have different values for the two nodes produced by the data_used_in_multiple_ways function -- public for column_a public private for column_b..

A note on decorator precedence. If using @tag together with @tag_outputs on a function (you might want to do this because you use @tag to "tag" all nodes with a certain set of values, and @tag_outputs to "tag" specific outputs with specific values), they will be applied in order up from the function. So if you desire to override @tag, for that to work, you would put tag_outputs above tag (as it would be applied last).

import pandas as pd
from hamilton.function_modifiers import tag_outputs, tag, extract_columns

def intermediate_column() -> pd.Series:
    pass

@tag_outputs(
    public={'column_a' : 'public'},
    private={'column_b' : 'private', 'common_tag' : 'bar'})
@tag(common_tag="foo")
@extract_columns('column_a', 'column_b')
def data_used_in_multiple_ways() -> pd.DataFrame:
    return load_some_data(...)

In the case above, common_tag would resolve to foo for column_a and bar for column_b. Attempting an override in the reverse direction is currently undefined behavior.

How do I query by tags?

Right now, we don't have a specific interface to query by tags, however we do expose them via the driver. Using the list_available_variables() capability exposes tags along with their names & types, enabling querying of the available outputs for specific tag matches. E.g.

from hamilton import driver
dr = driver.Driver(...)  # create driver as required
all_possible_outputs = dr.list_available_variables()
desired_outputs = [o.name for o in all_possible_outputs
                   if 'my_tag_value' == o.tags.get('my_tag_key')]
output = dr.execute(desired_outputs)

@check_output

The @check_output decorator enables you to add simple data quality checks to your code.

For example:

import pandas as pd
import numpy as np
from hamilton.function_modifiers import check_output

@check_output(
    data_type=np.int64,
    data_in_range=(0,100),
)
def some_int_data_between_0_and_100() -> pd.Series:
    pass

The check_output validator takes in arguments that each correspond to one of the default validators. These arguments tell it to add the default validator to the list. The above thus creates two validators, one that checks the datatype of the series, and one that checks whether the data is in a certain range.

Note that you can also specify custom decorators using the @check_output_custom decorator. check_output and check_output_custom accept the _target parameter, which specifies which output to check.

See data_quality for more information on available validators and how to build custom ones.

@subdag

The @subdag decorator enables you to rerun components of your DAG with varying parameters. Note that this is immensely powerful -- if we draw analogies from Hamilton to standard procedural programming paradigms, we might have the following correspondence:

  • config.when + friends -- if/else statements
  • parameterize/extract_columns -- for loop
  • does -- effectively macros And so on. subdag takes this one step further.
  • subdag -- subroutine definition E.G. take a certain set of nodes, and run them with specified parameters.

Why might you want to use this? Let's take a look at some examples:

  1. You have a feature engineering pipeline that you want to run on multiple datasets. If its exactly the same, this is perfect. If not, this works perfectly as well, you just have to utilize different functions in each or the config.when + config parameter to rerun it.
  2. You want to train multiple models in the same DAG that share some logic (in features or training) -- this allows you to reuse and continually add more.
  3. You want to combine multiple similar DAGs (e.g. one for each business line) into one so you can build a cross-business line model.

This basically bridges the gap between the flexibility of non-declarative pipelining frameworks with the readability/maintainability of declarative ones.

Let's take a look at a simplified example (in examples/).

@extract_columns("timestamp", "user_id", "region")  # one of "US", "CA" (canada)
def website_interactions() -> pd.DataFrame:
    return ...

def interactions_filtered(website_interactions: pd.DataFrame, region: str) -> pd.DataFrame:
    """Filters interactions by region -- note this will be run differently depending on the region its in"""
    pass

def unique_users(filtered_interactions: pd.DataFrame, grain: str) -> pd.Series:
    """Gives the number of shares traded by the frequency"""
    return ...

@subdag(
    unique_users, interactions_filtered,
    inputs={"grain": value("day")},
    config={"region": "US"},
)
def daily_users_US(unique_users: pd.Series) -> pd.Series:
    """Calculates quarterly data for just US users
    Note that this just returns the output. You can modify it if you'd like!
    """

    return unique_users


@subdag(
    unique_users, interactions_filtered,
    inputs={"grain": value("day")},
    config={"region": "CA"},
)
def daily_users_CA(unique_users: pd.Series) -> pd.Series:
    """Calculates quarterly data for just Canada users"""
    return unique_users

This example tracks users on a per-value user data. Specifically, we track the following:

  1. Daily user data for Canada
  2. Quarterly user data for the US

These each live under a separate namespace -- this exists solely so the two sets of similar nodes can coexist. Note this set is contrived to demonstrate functionality -- it should be easy to imagine how we could add more variations.

The inputs to the subdag decorator takes in a variety of inputs that determine which possible set of functions will constitute the DAG, and then specifically what the DAG is and how it is connected.

  • which functions constitute the subdag is specified by the *args input, which is a collection of modules/functions. These are used to determine the functions (i.e. nodes) that could end up in the produced subDAG.
  • what the sub-DAG is, and how it is connected is specified by two parameters. config provides configuration to use to generate the subDAG (in this case the region), and inputs provides inputs to connect the subdag with the current DAG (similar in spirit to @parameterize).

Note that, if you wanted to do this functionality without this decorator, you'd have two options:

  1. Rewrite every function for each scenario -- this is repetetive and doesn't scale
  2. Utilize the driver within the functions -- E.G.
def daily_users_CA(unique_users: pd.Series) -> pd.Series:
    """Calculates quarterly data for just canada users"""
    dr = hamilton.driver.Driver({"region" : "CA"}, unique_users, interactions_filtered)
    return dr.execute(["unique_users"], inputs={"grain": value("day")})["unique_users"]

While this is a clever approach (and you can use it if you want), there are a few drawbacks:

  1. You have no visibility into the subdag -- hamilton has no knowledge about what's being run
  2. Any execution parameters have to be reproduced for the driver. E.G. running on dask, etc..., is unknown to the driver.
  3. DAG compilation is done at runtime -- you would not know any type-errors until the function is running

That said, this is a good mental model for explaining subdag functionality. Think about it as a driver within a function, but in a way that hamilton can understand and operate. Best of both worlds!

@parameterize_extract_columns

@parameterize_extract_columns gives you the power of both @extract_columns and @parameterize in one decorator.

It takes in a list of Parameterized_Extract objects, each of which is composed of:

  1. A list of columns to extract, and
  2. A parameterization that gets used

In the following case, we produce four columns, two for each parameterization.

import pandas as pd
from function_modifiers import parameterize_extract_columns, ParameterizedExtract, source, value
@parameterize_extract_columns(
    ParameterizedExtract(
        ("outseries1a", "outseries2a"),
        {"input1": source("inseries1a"), "input2": source("inseries1b"), "input3": value(10)},
    ),
    ParameterizedExtract(
        ("outseries1b", "outseries2b"),
        {"input1": source("inseries2a"), "input2": source("inseries2b"), "input3": value(100)},
    ),
)
def fn(input1: pd.Series, input2: pd.Series, input3: float) -> pd.DataFrame:
    return pd.concat([input1 * input2 * input3, input1 + input2 + input3], axis=1)

@parameterize_frame

@parameterize_frame enables you to run parameterize_extract_columns with a dataframe specifying the parameterizations -- allowing for less verbose specification. The above example can be rewritten as:

from hamilton.experimental.decorators.parameterize_frame import parameterize_frame
df = pd.DataFrame(
        [
            ["outseries1a", "outseries2a", "inseries1a", "inseries2a", 10],
            ["outseries1b", "outseries2b", "inseries1b", "inseries2b", 100],
            # ...
        ],
        # Have to switch as indices have to be unique
        columns=[
            [
                "output1",
                "output2",
                "input1",
                "input2",
                "input3",
            ],  # configure whether column is source or value and also whether it's input ("source", "value") or output ("out")
            ["out", "out", "source", "source", "value"],
        ],
    )

@parameterize_frame(df)
def my_func(input1: pd.Series, input2: pd.Series, input3: float) -> pd.DataFrame:
    return pd.DataFrame(
        [input1 * input2 * input3, input1 + input2 + input3]
    )

Note that we have a double-index. Note that this is still in experimental, and has the possibility of being changed; we'd love feedback on this API if you end up using it!

@model

@model allows you to abstract a function that is a model. You will need to implement models that make sense for your business case. Reach out if you need examples.

Under the hood, they're just DAG nodes whose inputs are determined by a configuration parameter. A model takes in two required parameters:

  1. The class it uses to run the model. If external to Stitch Fix you will need to write your own, else internally see the internal docs for this. Basically the class defined determines what the function actually does.
  2. The configuration key that determines how the model functions. This is just the name of a configuration parameter that stores the way the model is run.

The following is an example usage of @model:

import pandas as pd
from hamilton.function_modifiers import model
import internal_package_with_logic

@model(internal_package_with_logic.GLM, 'model_p_cancel_manual_res')
# This runs a GLM (Generalized Linear Model)
# The associated configuration parameter is 'model_p_cancel_manual_res',
# which points to the results of loading the model_p_cancel_manual_res table
def prob_cancel_manual_res() -> pd.Series:
    pass

GLM here is not part of the hamilton framework, and instead a user defined model.

Models (optionally) accept a output_column parameter -- this is specifically if the name of the function differs from the output column that it should represent. E.G. if you use the model result as an intermediate object, and manipulate it all later. At Stitch Fix this is necessary because various dependent columns that a model queries (e.g. MULTIPLIER_... and OFFSET_...) are derived from the model's name.

A mental model for decorators

Decorators fall under a few broad categories:

  1. Decorators that decide whether a function should resolves into a node (E.G. config.when())
  2. Decorators that turn a function into a set of nodes (E.G. @does)
  3. Decorators that modify a set of nodes, turning it into another set of nodes (E.G. @parameterize, @extract_columns, and @check_output)

Note the distinction between "functions" and "nodes". "functions" are what the user writes, and in the plain case, correspond 1:1 to nodes. Nodes are then generated from functions -- decorators control how that happens. Thus, one function can correspond to 0, 1 or many nodes.

The decorators that do this are broken up into subclasses that each have various nuances, one of which is whether or not they allow layering. To layer effectively, we need to know which nodes a decorator will modify from the result of another decorator. For the ones that allow this type of layers, they take in a target_ parameter (with _ due to the fact that they often have **kwargs) that tells which nodes they should modify.

There are 4 options for target_:

  1. None -- this is the default behavior. This only applies to nodes generated by the function that are not depended on by any other nodes generated by the function.
  2. ... (the literal python Ellipsis). This means that all nodes generated by the function and its decorators will have the transformation applied to them.
  3. A string. This means that only the node with the specified name will have the transformation applied.
  4. A list of strings. This means that all nodes with the specified name will have the transformation applied.

This is very powerful, and we will likely be applying it to more decorators. Currently tag and check_output are the only decorators that accept this parameter.

To dive into this, let's take a look at the tag decorator with the default target_:

This will only decorate col1, col2, and col3:

import pandas as pd
from hamilton.function_modifiers import tag, extract_columns

@tag(type="dataframe")
@extract_columns("col1", "col2", "col3")
def dataframe() -> pd.DataFrame:
    pass

Once we decorate it with target_="original_dataframe", we are now just decorating the dataframe that gets extracted from (that is a node after all).

import pandas as pd
from hamilton.function_modifiers import tag, extract_columns

@tag(type="dataframe", target_='original_dataframe')
@extract_columns("col1", "col2", "col3")
def dataframe() -> pd.DataFrame:
    pass

Whereas the following would tag all the columns we extract:

import pandas as pd
from hamilton.function_modifiers import tag, extract_columns

@tag(type="extracted_column", target_=["col1", "col2", "col3"])
@extract_columns("col1", "col2", "col3")
def dataframe() -> pd.DataFrame:
    pass

The following would tag everything

import pandas as pd
from hamilton.function_modifiers import tag, extract_columns

@tag(type="any_node_created", target_=...)
@extract_columns("col1", "col2", "col3")
def dataframe() -> pd.DataFrame:
    pass

Passing None to target_ is the same as not passing it at all -- this decorates the "final" nodes as we saw in the first example.