Basic question in tabular

fastai fastai dev
1

I want to better understand how the dataloaders are actually passing data to the model… To better understand I thought I would simply create a batch from tabular data dataloader, pass this to a TabularModel, set some breakpoints in there. But I am having some problems.

results_tab = TabularPandas(results, [Categorify, FillMissing, Normalize], results_cat_names, results_cont_names, y_names='NextScore')
rdls = results_tab.dataloaders(bs=16)
results_model = TabularModel(results_emb_szs, len(results_cont_names), 100, [1000, 250], ps=[0.01, 0.1], embed_p=0.04).cuda()

results_model(rdls.one_batch())

The code crash while calling the model:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-88-3c075dd5c894> in <module>
      3 results_model = TabularModel(results_emb_szs, len(results_cont_names), 100, [1000, 250], ps=[0.01, 0.1], embed_p=0.04).cuda()
      4 
----> 5 results_model(rdls.one_batch())

~\Anaconda3\envs\fastai2\lib\site-packages\torch\nn\modules\module.py in __call__(self, *input, **kwargs)
    530             result = self._slow_forward(*input, **kwargs)
    531         else:
--> 532             result = self.forward(*input, **kwargs)
    533         for hook in self._forward_hooks.values():
    534             hook_result = hook(self, input, result)

c:\work\ml\fastai2\fastai2\tabular\model.py in forward(self, x_cat, x_cont)
     46     def forward(self, x_cat, x_cont=None):
     47         if self.n_emb != 0:
---> 48             x = [e(x_cat[:,i]) for i,e in enumerate(self.embeds)]
     49             x = torch.cat(x, 1)
     50             x = self.emb_drop(x)

c:\work\ml\fastai2\fastai2\tabular\model.py in <listcomp>(.0)
     46     def forward(self, x_cat, x_cont=None):
     47         if self.n_emb != 0:
---> 48             x = [e(x_cat[:,i]) for i,e in enumerate(self.embeds)]
     49             x = torch.cat(x, 1)
     50             x = self.emb_drop(x)

TypeError: tuple indices must be integers or slices, not tuple

Looking at the returned value from one_batch() I can see that this is a tuple with 3 things, (the categorical variables, the continuous variables, the targets). The model expects only the categorical variables and the continuous variables…

What am I doing wrong?

Thanks,

2

Ok I found my answer by looking at the Learner code…

I had to change the line calling the model to:

results_model(*rdls.one_batch()[:-1])

1 Like
3

Yes your model can never take the whole batch, and in this case, there are two inputs :slight_smile: