Defining dependencies and products#
Tasks have dependencies and products that you must define to run your tasks.
Defining dependencies and products also serves another purpose. By analyzing them, pytask determines the order to run the tasks.
This tutorial offers you different interfaces. If you are comfortable with type
annotations or are not afraid to try them, look at the Python 3.10+
or 3.8+
tabs.
You find a tutorial on type hints here.
If you want to avoid type annotations for now, look at the tab named produces
.
Warning
The Decorators
tab documents the deprecated approach that should not be used anymore
and will be removed in version v0.5.
See also
In this tutorial, we only deal with local files. If you want to use pytask with files online, S3, GCP, Azure, etc., read the guide on remote files.
First, we focus on defining products that should already be familiar to you. Then, we focus on how you can declare task dependencies.
We use the same project as before and add a task_plot_data.py
module.
my_project
│
├───.pytask
│
├───bld
│ ├────data.pkl
│ └────plot.png
│
├───src
│ └───my_project
│ ├────__init__.py
│ ├────config.py
│ ├────task_data_preparation.py
│ └────task_plot_data.py
│
└───pyproject.toml
Products#
Let’s revisit the task from the previous tutorial that we defined
in task_data_preparation.py
.
from pathlib import Path
from typing import Annotated
import numpy as np
import pandas as pd
from my_project.config import BLD
from pytask import Product
def task_create_random_data(
path_to_data: Annotated[Path, Product] = BLD / "data.pkl",
) -> None:
rng = np.random.default_rng(0)
beta = 2
x = rng.normal(loc=5, scale=10, size=1_000)
epsilon = rng.standard_normal(1_000)
y = beta * x + epsilon
df = pd.DataFrame({"x": x, "y": y})
df.to_pickle(path_to_data)
Product
allows marking an argument as a product. After the
task has finished, pytask will check whether the file exists.
from pathlib import Path
import numpy as np
import pandas as pd
from my_project.config import BLD
from pytask import Product
from typing_extensions import Annotated
def task_create_random_data(
path_to_data: Annotated[Path, Product] = BLD / "data.pkl",
) -> None:
rng = np.random.default_rng(0)
beta = 2
x = rng.normal(loc=5, scale=10, size=1_000)
epsilon = rng.standard_normal(1_000)
y = beta * x + epsilon
df = pd.DataFrame({"x": x, "y": y})
df.to_pickle(path_to_data)
Product
allows marking an argument as a product. After the
task has finished, pytask will check whether the file exists.
from pathlib import Path
import numpy as np
import pandas as pd
from my_project.config import BLD
def task_create_random_data(produces: Path = BLD / "data.pkl") -> None:
rng = np.random.default_rng(0)
beta = 2
x = rng.normal(loc=5, scale=10, size=1_000)
epsilon = rng.standard_normal(1_000)
y = beta * x + epsilon
df = pd.DataFrame({"x": x, "y": y})
df.to_pickle(produces)
Tasks can use produces
as a “magic” argument name. Every value, or in this case path,
passed to this argument is automatically treated as a task product. Here, we pass the
path as the default argument.
Warning
This approach is deprecated and will be removed in v0.5
from pathlib import Path
import numpy as np
import pandas as pd
import pytask
from my_project.config import BLD
@pytask.mark.produces(BLD / "data.pkl")
def task_create_random_data(produces: Path) -> None:
rng = np.random.default_rng(0)
beta = 2
x = rng.normal(loc=5, scale=10, size=1_000)
epsilon = rng.standard_normal(1_000)
y = beta * x + epsilon
df = pd.DataFrame({"x": x, "y": y})
df.to_pickle(produces)
The @pytask.mark.produces
marker attaches a product to a
task. After the task has finished, pytask will check whether the file exists.
Add produces
as an argument of the task function to get access to the same path inside
the task function.
Tip
If you do not know about pathlib
check out this guide by
RealPython. The module is beneficial for
handling paths conveniently and across platforms.
Dependencies#
Adding a dependency to a task ensures that the dependency is available before execution.
To show how dependencies work, we extend our project with another task that plots the
data generated with task_create_random_data
. The task is called task_plot_data
, and
we will define it in task_plot_data.py
.
To specify that the task relies on the data set data.pkl
, you can add the path
to the function signature while choosing any argument name, here path_to_data
.
pytask assumes that all function arguments that do not have a Product
annotation are dependencies of the task.
from pathlib import Path
from typing import Annotated
import matplotlib.pyplot as plt
import pandas as pd
from my_project.config import BLD
from pytask import Product
def task_plot_data(
path_to_data: Path = BLD / "data.pkl",
path_to_plot: Annotated[Path, Product] = BLD / "plot.png",
) -> None:
df = pd.read_pickle(path_to_data)
_, ax = plt.subplots()
df.plot(x="x", y="y", ax=ax, kind="scatter")
plt.savefig(path_to_plot)
plt.close()
To specify that the task relies on the data set data.pkl
, you can add the path
to the function signature while choosing any argument name, here path_to_data
.
pytask assumes that all function arguments that do not have the Product
annotation are dependencies of the task.
from pathlib import Path
import matplotlib.pyplot as plt
import pandas as pd
from my_project.config import BLD
from pytask import Product
from typing_extensions import Annotated
def task_plot_data(
path_to_data: Path = BLD / "data.pkl",
path_to_plot: Annotated[Path, Product] = BLD / "plot.png",
) -> None:
df = pd.read_pickle(path_to_data)
_, ax = plt.subplots()
df.plot(x="x", y="y", ax=ax, kind="scatter")
plt.savefig(path_to_plot)
plt.close()
To specify that the task relies on the data set data.pkl
, you can add the path to the
function signature while choosing any argument name, here path_to_data
.
pytask assumes that all function arguments that are not passed to the argument
produces
are dependencies of the task.
from pathlib import Path
import matplotlib.pyplot as plt
import pandas as pd
from my_project.config import BLD
def task_plot_data(
path_to_data: Path = BLD / "data.pkl", produces: Path = BLD / "plot.png"
) -> None:
df = pd.read_pickle(path_to_data)
_, ax = plt.subplots()
df.plot(x="x", y="y", ax=ax, kind="scatter")
plt.savefig(produces)
plt.close()
Warning
This approach is deprecated and will be removed in v0.5
Equivalent to products, you can use the
@pytask.mark.depends_on
decorator to specify that
data.pkl
is a dependency of the task. Use depends_on
as a function argument to
access the dependency path inside the function and load the data.
from pathlib import Path
import matplotlib.pyplot as plt
import pandas as pd
import pytask
from my_project.config import BLD
@pytask.mark.depends_on(BLD / "data.pkl")
@pytask.mark.produces(BLD / "plot.png")
def task_plot_data(depends_on: Path, produces: Path) -> None:
df = pd.read_pickle(depends_on)
_, ax = plt.subplots()
df.plot(x="x", y="y", ax=ax, kind="scatter")
plt.savefig(produces)
plt.close()
Now, let us execute the two paths.
$ pytask
──────────────────────────── Start pytask session ────────────────────────────
Platform: win32 -- Python <span style="color: var(--termynal-blue)">3.10.0</span>, pytask <span style="color: var(--termynal-blue)">0.4.0</span>, pluggy <span style="color: var(--termynal-blue)">1.0.0</span>
Root: C:\Users\pytask-dev\git\my_project
Collected <span style="color: var(--termynal-blue)">2</span> task.
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━┓
┃ Task ┃ Outcome ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━┩
│ <span class="termynal-dim">task_data_preparation.py::</span>task_create_random_data │ <span class="termynal-success">.</span> │
│ <span class="termynal-dim">task_plot_data.py::</span>task_plot_data │ <span class="termynal-success">.</span> │
└───────────────────────────────────────────────────┴─────────┘
<span class="termynal-dim">──────────────────────────────────────────────────────────────────────────────</span>
<span class="termynal-success">╭───────────</span> <span style="font-weight: bold;">Summary</span> <span class="termynal-success">────────────╮</span>
<span class="termynal-success">│</span> <span style="font-weight: bold;"> 2 Collected tasks </span> <span class="termynal-success">│</span>
<span class="termynal-success">│</span> <span class="termynal-success-textonly"> 2 Succeeded (100.0%) </span> <span class="termynal-success">│</span>
<span class="termynal-success">╰────────────────────────────────╯</span>
<span class="termynal-success">───────────────────────── Succeeded in 0.06 seconds ──────────────────────────</span>
Relative paths#
Dependencies and products do not have to be absolute paths. If paths are relative, they are assumed to point to a location relative to the task module.
from pathlib import Path
from typing import Annotated
from pytask import Product
def task_create_random_data(
path_to_data: Annotated[Path, Product] = Path("../bld/data.pkl"),
) -> None:
...
from pathlib import Path
from pytask import Product
from typing_extensions import Annotated
def task_create_random_data(
path_to_data: Annotated[Path, Product] = Path("../bld/data.pkl"),
) -> None:
...
from pathlib import Path
def task_create_random_data(produces: Path = Path("../bld/data.pkl")) -> None:
...
Warning
This approach is deprecated and will be removed in v0.5
You can also use absolute and relative paths as strings that obey the same rules as the
pathlib.Path
.
from pathlib import Path
import pytask
@pytask.mark.produces("../bld/data.pkl")
def task_create_random_data(produces: Path) -> None:
...
If you use depends_on
or produces
as arguments for the task function, you will have
access to the paths of the targets as pathlib.Path
.
Multiple dependencies and products#
Of course, tasks can have multiple dependencies and products.
from pathlib import Path
from typing import Annotated
from my_project.config import BLD
from pytask import Product
def task_plot_data(
path_to_data_0: Path = BLD / "data_0.pkl",
path_to_data_1: Path = BLD / "data_1.pkl",
path_to_plot_0: Annotated[Path, Product] = BLD / "plot_0.png",
path_to_plot_1: Annotated[Path, Product] = BLD / "plot_1.png",
) -> None:
...
You can group your dependencies and product if you prefer not to have a function argument per input. Use dictionaries (recommended), tuples, lists, or more nested structures if needed.
from pathlib import Path
from typing import Annotated
from my_project.config import BLD
from pytask import Product
_DEPENDENCIES = {"data_0": BLD / "data_0.pkl", "data_1": BLD / "data_1.pkl"}
_PRODUCTS = {"plot_0": BLD / "plot_0.png", "plot_1": BLD / "plot_1.png"}
def task_plot_data(
path_to_data: dict[str, Path] = _DEPENDENCIES,
path_to_plots: Annotated[dict[str, Path], Product] = _PRODUCTS,
) -> None:
...
from pathlib import Path
from my_project.config import BLD
from pytask import Product
from typing_extensions import Annotated
def task_plot_data(
path_to_data_0: Path = BLD / "data_0.pkl",
path_to_data_1: Path = BLD / "data_1.pkl",
path_to_plot_0: Annotated[Path, Product] = BLD / "plot_0.png",
path_to_plot_1: Annotated[Path, Product] = BLD / "plot_1.png",
) -> None:
...
You can group your dependencies and product if you prefer not to have a function argument per input. Use dictionaries (recommended), tuples, lists, or more nested structures if needed.
from pathlib import Path
from typing import Dict
from my_project.config import BLD
from pytask import Product
from typing_extensions import Annotated
_DEPENDENCIES = {"data_0": BLD / "data_0.pkl", "data_1": BLD / "data_1.pkl"}
_PRODUCTS = {"plot_0": BLD / "plot_0.png", "plot_1": BLD / "plot_1.png"}
def task_plot_data(
path_to_data: Dict[str, Path] = _DEPENDENCIES,
path_to_plots: Annotated[Dict[str, Path], Product] = _PRODUCTS,
) -> None:
...
If your task has multiple products, group them in one container like a dictionary (recommended), tuples, lists, or more nested structures.
from pathlib import Path
from typing import Dict
from my_project.config import BLD
_PRODUCTS = {"first": BLD / "data_0.pkl", "second": BLD / "data_1.pkl"}
def task_plot_data(
path_to_data_0: Path = BLD / "data_0.pkl",
path_to_data_1: Path = BLD / "data_1.pkl",
produces: Dict[str, Path] = _PRODUCTS,
) -> None:
...
You can do the same with dependencies.
from __future__ import annotations
from typing import TYPE_CHECKING
from my_project.config import BLD
if TYPE_CHECKING:
from pathlib import Path
_DEPENDENCIES = {"data_0": BLD / "data_0.pkl", "data_1": BLD / "data_1.pkl"}
_PRODUCTS = {"plot_0": BLD / "plot_0.png", "plot_1": BLD / "plot_1.png"}
def task_plot_data(
path_to_data: dict[str, Path] = _DEPENDENCIES,
produces: dict[str, Path] = _PRODUCTS,
) -> None:
...
Warning
This approach is deprecated and will be removed in v0.5
The easiest way to attach multiple dependencies or products to a task is to pass a
dict
(highly recommended), list
, or another iterator to the marker
containing the paths.
To assign labels to dependencies or products, pass a dictionary. For example,
from typing import Dict
@pytask.mark.produces({"first": BLD / "data_0.pkl", "second": BLD / "data_1.pkl"})
def task_create_random_data(produces: Dict[str, Path]) -> None:
...
Then, use produces
inside the task function.
>>> produces["first"]
BLD / "data_0.pkl"
>>> produces["second"]
BLD / "data_1.pkl"
You can also use lists and other iterables.
@pytask.mark.produces([BLD / "data_0.pkl", BLD / "data_1.pkl"])
def task_create_random_data(produces):
...
Inside the function, the arguments depends_on
or produces
become a dictionary where
keys are the positions in the list.
>>> produces
{0: BLD / "data_0.pkl", 1: BLD / "data_1.pkl"}
Why does pytask recommend dictionaries and convert lists, tuples, or other iterators to dictionaries? First, dictionaries with positions as keys behave very similarly to lists.
Secondly, dictionary keys are more descriptive and do not assume a fixed ordering. Both attributes are especially desirable in complex projects.
Multiple decorators
pytask merges multiple decorators of one kind into a single dictionary. This might help you to group dependencies and apply them to multiple tasks.
common_dependencies = pytask.mark.depends_on(
{"first_text": "text_1.txt", "second_text": "text_2.txt"}
)
@common_dependencies
@pytask.mark.depends_on("text_3.txt")
def task_example(depends_on):
...
Inside the task, depends_on
will be
>>> depends_on
{"first_text": ... / "text_1.txt", "second_text": "text_2.txt", 0: "text_3.txt"}
Nested dependencies and products
Dependencies and products can be nested containers consisting of tuples, lists, and dictionaries. It is beneficial if you want more structure and nesting.
Here is an example of a task that fits some model on data. It depends on a module containing the code for the model, which is not actively used but ensures that the task is rerun when the model is changed. And it depends on the data.
@pytask.mark.depends_on(
{
"model": [SRC / "models" / "model.py"],
"data": {"a": SRC / "data" / "a.pkl", "b": SRC / "data" / "b.pkl"},
}
)
@pytask.mark.produces(BLD / "models" / "fitted_model.pkl")
def task_fit_model(depends_on, produces):
...
depends_on
within the function will be
{
"model": [SRC / "models" / "model.py"],
"data": {"a": SRC / "data" / "a.pkl", "b": SRC / "data" / "b.pkl"},
}
Depending on a task#
In some situations, you want to define a task depending on another task.
pytask allows you to do that, but you lose features like access to paths, which is why defining dependencies explicitly is always preferred.
There are two modes for it, and both use @task(after=...)
.
First, you can pass the task function or multiple task functions to the decorator.
Applied to the tasks from before, we could have written task_plot_data
as
@task(after=task_create_random_data)
def task_plot_data(...):
...
You can also pass a list of task functions.
The second mode is to pass an expression, a substring of the name of the dependent tasks. Here, we can pass the function name or a significant part of the function name.
@task(after="random_data")
def task_plot_data(...):
...
You will learn more about expressions in Selecting tasks.
Further reading#
There is an additional way to specify products by treating the returns of a task function as a product. Read Using task returns to learn more about it.
An overview of all ways to specify dependencies and products and their strengths and weaknesses can be found in Interfaces for dependencies and products.