Putting Series in Project

[MarieS-WiMLDS]

I would like to be able to store series objects in project.

Here is a piece of code that doesn't work for now:

# %%
import pandas as pd
from skore import load

# %%
# !python -m skore create

# %%
project = load("project.skore")

# %%
my_df = pd.Series(data=[1,2,3])

#%%
type(my_df)
# %%
project.put("my_df", my_df)

I get:

{
	"name": "AttributeError",
	"message": "module 'matplotlib' has no attribute 'figure'",
	"stack": "---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
File /home/marie/Documents/ML_use_cases/binary_class_tabular/test.py:2
      1 # %%
----> 2 project.put(\"my_df\", my_df)

File ~/anaconda3/envs/skore/lib/python3.12/site-packages/skore/project.py:57, in Project.put(self, key, value)
     55 def put(self, key: str, value: Any):
     56     \"\"\"Add a value to the Project.\"\"\"
---> 57     item = object_to_item(value)
     58     self.put_item(key, item)

File ~/anaconda3/envs/skore/lib/python3.12/site-packages/skore/project.py:36, in object_to_item(o)
     34 elif isinstance(o, altair.vegalite.v5.schema.core.TopLevelSpec):
     35     return MediaItem.factory_altair(o)
---> 36 elif isinstance(o, matplotlib.figure.Figure):
     37     return MediaItem.factory_matplotlib(o)
     38 elif isinstance(o, PIL.Image.Image):

File ~/anaconda3/envs/skore/lib/python3.12/site-packages/matplotlib/_api/__init__.py:217, in caching_module_getattr.<locals>.__getattr__(name)
    215 if name in props:
    216     return props[name].__get__(instance)
--> 217 raise AttributeError(
    218     f\"module {cls.__module__!r} has no attribute {name!r}\")

AttributeError: module 'matplotlib' has no attribute 'figure'"

Thanks!

[augustebaum]

The traceback you're getting is a bug unrelated to our ability to store Series; it was fixed by commit 27cf3c5 and is part of the new pre-release 0.0.1rc3

[augustebaum]

Here is the error I get with the newer version:

---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
Cell In[4], line 1
----> 1 p.put("he", pd.Series([1,2]))

File ~/Desktop/skore/src/skore/project.py:50, in Project.put(self, key, value)
     48 def put(self, key: str, value: Any):
     49     """Add a value to the Project."""
---> 50     item = object_to_item(value)
     51     self.put_item(key, item)

File ~/Desktop/skore/src/skore/project.py:34, in object_to_item(o)
     32     return MediaItem.factory_pillow(o)
     33 else:
---> 34     raise NotImplementedError(f"Type {o.__class__.__name__} is not supported yet.")

NotImplementedError: Type Series is not supported yet.

[MarieS-WiMLDS]

it seems more logical indeed!

[rouk1]

Open question: what would be a good widget to visualize a series ? A plot ?
I see that pandas can build a plot from a series btw.

[MarieS-WiMLDS]

That's a very good question. I would expect it to be displayed like it is python (see examples here). It's not pretty but I have no other idea I must say.

[thomass-dev]

@MarieS-WiMLDS in your workflow, do you need to retrieve a series from the project and use it in your script?

[tuscland]

@thomass-dev do you agree that this message is kind of confusing:

Type Series is not supported yet
It suggests that we might support it later.

I propose to remove the ending "yet": "Type Series is not supported."

[MarieS-WiMLDS]

@MarieS-WiMLDS in your workflow, do you need to retrieve a series from the project and use it in your script?

yes, i started to use skore also has a link between notebooks when I have several of them in my project. Here is why I want to store series but didn't think about displaying them, as I don't really know how it could be useful in a report.

[thomass-dev]

Is it acceptable to say: series can be transformed into a list or created from a list.
If the user wants to set persistent series, he can convert the series into a list himself.

project.put("myserie", myserie.to_list())
[...]
myserie = project.put("myserie", myserie.to_list())
myserie = pandas.Series(myseries)

What do you think ?
I'm just afraid that we should support many many types based on list: pandas.Series, pandas.Index etc.

[MarieS-WiMLDS]

Yesterday when I saw your comment I thought that you were right and indeed we don't want to have to support many types.
Then I thought about it overnight, and now I think that Series are very important in a data science project, almost always the target in classical projects are stored as series. Could we add it to the short list of data types we will maintain?

[augustebaum]

I propose to remove the ending "yet": "Type Series is not supported."

This is done

[tuscland]

How can @MarieS-WiMLDS test it?
In this issue, there is no reference to a PR or a branch.

[thomass-dev]

The comment of @augustebaum was only on the renaming of the message exception, i.e. "yet".
It was done before your comment, here in #355 .