wandb mistakes integers and dictionaries as DataFrames?

wandb --version && python --version && uname I’m running Wandb locally via wandb local It is running successfully on localhost:8080

• Weights and Biases version: wandb, version 0.9.7
• Python version: Python 3.7.9
• Operating System: Windows 10

Description

Describe what you were trying to get done. Just simple config logging of hyperparameters.

I set wandb.config.verbosity = 3 and wandb.config.hyperparameters = {'NN': {'num_epochs': 5}} to record the verbosity and hyperparameters of a training model with the AutoGluon framework.

Tell us what happened, what went wrong, and what you expected to happen.

wandb.config.tags = [col for col in columns_to_keep]  # use the feature column names as tags 
wandb.config.metric_name = 'roc_auc'
wandb.config.trial_datetime = utils.get_current_time()

print(f'wandb.config.trial_datetime is :   {wandb.config.trial_datetime}\n')

wandb.config.output_directory = str(Path(config.AUTOGLUON_MODELS_DIR, wandb.config.trial_datetime))
wandb.config.eval_metric = wandb.config.metric_name
wandb.config.auto_stack = True
wandb.config.num_trials = 5
wandb.config.time_limits = config.MAX_RUNTIME_SECS  #1*60 #
wandb.config.num_bagging_folds = config.FOLDS # 5
wandb.config.stack_ensemble_levels = 3
wandb.config.visualizer = 'tensorboard'
wandb.config.verbosity = 3
wandb.config.hyperparameter_config_option = 'default'
wandb.config.hyperparameter_config_dict = get_hyperparameter_config(wandb.config.hyperparameter_config_option)

if train_with_neural_network == True:
    wandb.config.hyperparameters = {
                                        'NN': {'num_epochs':5}
                                        # 'GBM': {'num_boost_round':100}
                                    }

After setting the wandb config values, I simply print them:

print(wandb.config.hyperparameters) print(wandb.config.verbosity)

Expected behavior is that the simple integer value of 3 and the dictionary {'NN': {'num_epochs': 5}} would be interpreted correctly as an integer and dictionary, respectively, and not as a DataFrame.

Actual behavior: prints value, then throws an error. Curiously, I was logging a bunch of values in wandb.config and had no issue until this one.

Also curiously, it does indeed print the config value, but then immediately prints the AssertionError.

AssertionError: We don’t yet support DataFrames in History. Use wandb.Table(dataframe=my_dataframe)

What I Did

wandb.config.hyperparameters = {'NN': {'num_epochs': 5}}
wandb.config.verbosity = 3

print(wandb.config.hyperparameters)

{'NN': {'num_epochs': 5}}

Error in callback <function _init_jupyter.<locals>.cleanup at 0x00000158AAC33558> (for post_run_cell):
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\backcall\backcall.py in adapted(*args, **kwargs)
    102                 kwargs.pop(name)
    103 #            print(args, kwargs, unmatched_pos, cut_positional, unmatched_kw)
--> 104             return callback(*args, **kwargs)
    105 
    106         return adapted

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\__init__.py in cleanup()
    528         # shutdown async logger because _user_process_finished isn't called in jupyter
    529         shutdown_async_log_thread()
--> 530         run._stop_jupyter_agent()
    531     if hasattr(ipython.events, "_orig_post_run"):
    532         ipython.events.unregister("post_run_cell", ipython.events._orig_post_run)

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\wandb_run.py in _stop_jupyter_agent(self)
    211 
    212     def _stop_jupyter_agent(self):
--> 213         self._jupyter_agent.stop()
    214 
    215     def send_message(self, options):

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\jupyter.py in stop(self)
    140             self.save_history()
    141             self.rm.unmirror_stdout_stderr()
--> 142             wandb.run.close_files()
    143             self.rm.shutdown()
    144             self.paused = True

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\wandb_run.py in close_files(self)
    910             self._events = None
    911         if self._history is not None:
--> 912             self._history.close()
    913             self._history = None
    914 

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\history.py in close(self)
    260 
    261     def close(self):
--> 262         self._write()
    263         self._lock.acquire()
    264         try:

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\history.py in _write(self)
    244                 if self.stream_name != "default":
    245                     self.row["_stream"] = self.stream_name
--> 246                 self._transform()
    247                 self._file.write(util.json_dumps_safer_history(self.row))
    248                 self._file.write('\n')

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\history.py in _transform(self)
    225     def _transform(self):
    226         """Transforms special classes into the proper format before writing"""
--> 227         self.row = data_types.history_dict_to_json(self._run, self.row)
    228 
    229     def _write(self):

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\data_types.py in history_dict_to_json(run, payload, step)
   1707             payload[key] = history_dict_to_json(run, val, step=step)
   1708         else:
-> 1709             payload[key] = val_to_json(run, key, val, step=step)
   1710 
   1711     return payload

C:\ProgramData\Anaconda3\envs\arn\lib\site-packages\wandb\data_types.py in val_to_json(run, key, val, step)
   1735 
   1736     if util.is_pandas_data_frame(val):
-> 1737         assert step == 'summary', "We don't yet support DataFrames in History. Use `wandb.Table(dataframe=my_dataframe)`"
   1738         return data_frame_to_json(val, run, key, step)
   1739     elif util.is_matplotlib_typename(typename) or util.is_plotly_typename(typename):

AssertionError: We don't yet support DataFrames in History. Use `wandb.Table(dataframe=my_dataframe)`

Why does WandB misinterpret an integer as a DataFrame??

[issue-label-bot[bot]]

Issue-Label Bot is automatically applying the label bug to this issue, with a confidence of 0.69. Please mark this comment with 👍 or 👎 to give our bot feedback!

Links: app homepage, dashboard and code for this bot.

[sydholl]

In the past year we’ve majorly reworked the CLI and UI for Weights & Biases. We’re closing issues older than 6 months. Please comment to reopen.