[CLI] Can't log Seaborn plots: "AttributeError: 'PathCollection' object has no attribute 'get_offset_position'"

Describe the bug

Can’t seem to log Seaborn plots through the wandb.log command, it crashes every time. (I did check the issue https://github.com/wandb/wandb/issues/1608)

The program fails with the following exception:

AttributeError: 'PathCollection' object has no attribute 'get_offset_position'

Code for testing:

import seaborn as sns
import matplotlib.pyplot as plt
import wandb


wandb.init("test_plotly")

tips = sns.load_dataset("tips")
fig, ax = plt.subplots(figsize=(11.7, 8.27))
sns.scatterplot(data=tips, x="total_bill", y="tip")
wandb.log({"plot":fig})

wandb.finish()

Full trace:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Input In [2], in <cell line: 11>()
      9 fig, ax = plt.subplots(figsize=(11.7, 8.27))
     10 sns.scatterplot(data=tips, x="total_bill", y="tip")
---> 11 wandb.log({"plot":fig})
     13 wandb.finish()

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/wandb_run.py:256, in _run_decorator._noop.<locals>.wrapper(self, *args, **kwargs)
    253         wandb.termwarn(message, repeat=False)
    254         return cls.Dummy()
--> 256 return func(self, *args, **kwargs)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/wandb_run.py:222, in _run_decorator._attach.<locals>.wrapper(self, *args, **kwargs)
    220         raise e
    221     cls._is_attaching = ""
--> 222 return func(self, *args, **kwargs)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/wandb_run.py:1543, in Run.log(self, data, step, commit, sync)
   1536 if sync is not None:
   1537     deprecate.deprecate(
   1538         field_name=deprecate.Deprecated.run__log_sync,
   1539         warning_message=(
   1540             "`sync` argument is deprecated and does not affect the behaviour of `wandb.log`"
   1541         ),
   1542     )
-> 1543 self._log(data=data, step=step, commit=commit)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/wandb_run.py:1334, in Run._log(self, data, step, commit)
   1331 if any(not isinstance(key, str) for key in data.keys()):
   1332     raise ValueError("Key values passed to `wandb.log` must be strings.")
-> 1334 self._partial_history_callback(data, step, commit)
   1336 if step is not None:
   1337     if os.getpid() != self._init_pid or self._is_attached:

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/wandb_run.py:1223, in Run._partial_history_callback(self, row, step, commit)
   1220 if self._backend and self._backend.interface:
   1221     not_using_tensorboard = len(wandb.patched["tensorboard"]) == 0
-> 1223     self._backend.interface.publish_partial_history(
   1224         row,
   1225         user_step=self._step,
   1226         step=step,
   1227         flush=commit,
   1228         publish_step=not_using_tensorboard,
   1229     )

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/interface/interface.py:541, in InterfaceBase.publish_partial_history(self, data, user_step, step, flush, publish_step, run)
    530 def publish_partial_history(
    531     self,
    532     data: dict,
   (...)
    537     run: Optional["Run"] = None,
    538 ) -> None:
    539     run = run or self._run
--> 541     data = history_dict_to_json(run, data, step=user_step, ignore_copy_err=True)
    542     data.pop("_step", None)
    544     partial_history = pb.PartialHistoryRequest()

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/data_types/utils.py:54, in history_dict_to_json(run, payload, step, ignore_copy_err)
     50         payload[key] = history_dict_to_json(
     51             run, val, step=step, ignore_copy_err=ignore_copy_err
     52         )
     53     else:
---> 54         payload[key] = val_to_json(
     55             run, key, val, namespace=step, ignore_copy_err=ignore_copy_err
     56         )
     58 return payload

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/data_types/utils.py:82, in val_to_json(run, key, val, namespace, ignore_copy_err)
     79     val = wandb.Table(dataframe=val)
     81 elif util.is_matplotlib_typename(typename) or util.is_plotly_typename(typename):
---> 82     val = Plotly.make_plot_media(val)
     83 elif isinstance(val, Sequence) and all(isinstance(v, WBValue) for v in val):
     84     assert run

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/sdk/data_types/plotly.py:48, in Plotly.make_plot_media(cls, val)
     46     if util.matplotlib_contains_images(val):
     47         return Image(val)
---> 48     val = util.matplotlib_to_plotly(val)
     49 return cls(val)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/wandb/util.py:565, in matplotlib_to_plotly(obj)
    557 obj = ensure_matplotlib_figure(obj)
    558 tools = get_module(
    559     "plotly.tools",
    560     required=(
   (...)
    563     ),
    564 )
--> 565 return tools.mpl_to_plotly(obj)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/plotly/tools.py:111, in mpl_to_plotly(fig, resize, strip_style, verbose)
    109 if matplotlylib:
    110     renderer = matplotlylib.PlotlyRenderer()
--> 111     matplotlylib.Exporter(renderer).run(fig)
    112     if resize:
    113         renderer.resize()

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/plotly/matplotlylib/mplexporter/exporter.py:53, in Exporter.run(self, fig)
     50     import matplotlib.pyplot as plt
     52     plt.close(fig)
---> 53 self.crawl_fig(fig)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/plotly/matplotlylib/mplexporter/exporter.py:124, in Exporter.crawl_fig(self, fig)
    122 with self.renderer.draw_figure(fig=fig, props=utils.get_figure_properties(fig)):
    123     for ax in fig.axes:
--> 124         self.crawl_ax(ax)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/plotly/matplotlylib/mplexporter/exporter.py:146, in Exporter.crawl_ax(self, ax)
    144     self.draw_patch(ax, patch)
    145 for collection in ax.collections:
--> 146     self.draw_collection(ax, collection)
    147 for image in ax.images:
    148     self.draw_image(ax, image)

File ~/.local/share/virtualenvs/data-augmentation-0AjDNs-r/lib/python3.10/site-packages/plotly/matplotlylib/mplexporter/exporter.py:289, in Exporter.draw_collection(self, ax, collection, force_pathtrans, force_offsettrans)
    280 styles = {
    281     "linewidth": collection.get_linewidths(),
    282     "facecolor": collection.get_facecolors(),
   (...)
    285     "zorder": collection.get_zorder(),
    286 }
    288 offset_dict = {"data": "before", "screen": "after"}
--> 289 offset_order = offset_dict[collection.get_offset_position()]
    292 self.renderer.draw_path_collection(
    293     paths=processed_paths,
    294     path_coordinates=path_coords,
   (...)
    300     mplobj=collection,
    301 )

AttributeError: 'PathCollection' object has no attribute 'get_offset_position'

Additional Files

No response

Environment

WandB version: 0.12.21

OS: Fedora 36

Python version: python 3.10.5

Versions of relevant libraries (simplified pipenv graph)

aiohttp==3.8.1
hydra-colorlog==1.2.0
hydra-optuna-sweeper==1.2.0
jupyterlab==3.4.4
lightning-bolts==0.5.0
more-itertools==8.13.0
pip==22.1.2
plotly==5.9.0
pre-commit==2.20.0
pyrootutils==1.0.4
pytest==7.1.2
rich==12.5.1
scikit-learn==1.1.1
seaborn==0.11.2
sh==1.14.3
torchvision==0.13.0
wandb==0.12.21

Additional Context

No response

[naumanjaved]

Until the above referenced plotly issue is fixed to enable compatibility with matplotlib > 3.5, one workaround is to just use matplotlib==3.4.3

[RafayAK]

This exact issue is mentioned here: https://github.com/plotly/plotly.py/issues/3624