Updating with empty rows

[avril-affine]

Typically during training a DL model, there are many train losses/metrics and less validation losses/metrics. In this case, there would not be many "validation" update rows. Here is an example:

In [1]: from cox.store import Store

In [2]: store = Store('test')
Logging in: /home/panda/test/05470312-67d5-405b-ae74-d81b3b770be4

In [3]: store.add_table('metrics', {'train': float, 'val': float}
   ...: )
Out[3]: <cox.store.Table at 0x7f759f9af0b8>

In [4]: store['metrics']
Out[4]: <cox.store.Table at 0x7f759f9af0b8>

In [5]: t = store['metrics']

In [6]: t.update_row({'train': 1.23})

In [7]: t.flush_row()
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
~/anaconda3/envs/dev/lib/python3.7/site-packages/cox/store.py in flush_row(self)
    390             try:
--> 391                 assert self._curr_row_data[k] is not None
    392             except:

AssertionError:

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-7-d818515cef82> in <module>
----> 1 t.flush_row()

~/anaconda3/envs/dev/lib/python3.7/site-packages/cox/store.py in flush_row(self)
    397                     msg = 'Col %s is None!' % k
    398
--> 399                 raise ValueError(msg)
    400
    401         for k, v in self._curr_row_data.items():

ValueError: Col val is None!

Would it be possible to add an optional argument to flush_row() to allow None or would that mess things up down the line?

[lengstrom]

Yes I like the option of adding an allow_none option, we'll add this feature soon!