feature_spec dnn example

[dfalbel]

@skeydan here's the current code for the feature_spec example

[skeydan]

great, thanks! I nearly would have clicked "merge" but we should probably merge rstudio/tfdatasets#42 first ;-)
Do you think we could do that now? In the example I'm getting

 Error in sort.int(x, na.last = na.last, decreasing = decreasing, ...) : 
  'x' must be atomic 
16.
stop("'x' must be atomic") 
15.
sort.int(x, na.last = na.last, decreasing = decreasing, ...) 
14.
sort.default(unique(c(self$vocabulary_list_aux, unq))) 
13.
sort(unique(c(self$vocabulary_list_aux, unq))) at feature_spec.R#578
12.
self$steps[[i]]$fit_batch(nxt) at feature_spec.R#198
11.
spec$fit() at feature_spec.R#984
10.
fit.FeatureSpec(.) 
9.
fit(.) 
8.
function_list[[k]](value) 
7.
withVisible(function_list[[k]](value)) 
6.
freduce(value, `_function_list`) 
5.
`_fseq`(`_lhs`) 
4.
eval(quote(`_fseq`(`_lhs`)), env, env) 
3.
eval(quote(`_fseq`(`_lhs`)), env, env) 
2.
withVisible(eval(quote(`_fseq`(`_lhs`)), env, env)) 
1.
training %>% select(-id) %>% feature_spec(target ~ .) %>% step_numeric_column(ends_with("bin")) %>% 
    step_numeric_column(-ends_with("bin"), -ends_with("cat"), 
        normalizer_fn = scaler_standard()) %>% step_categorical_column_with_vocabulary_list(ends_with("cat")) %>% 
    step_embedding_column(ends_with("cat"), dimension = function(vocab_size) as.integer(sqrt(vocab_size) +  ... 

... perhaps because I've not updated tfdatasets from the PR?

[skeydan]

Actually one suggestion...

You know what, I had started thinking about this (which hadn't been there before)

porto <- porto %>%
  mutate_at(vars(ends_with("cat")), as.character)

and so at first I was thinking, ok, we want to tell the "text embeddings story" so yeah, it's a bit artificial but why not ... but now I'm back to what had been my very first reaction on seeing the example, namely, that texts normally have more than one word anyway and that 's what people will want to see?

What do you think about we use

step_indicator_column(ends_with("cat")) %>%`

and frame the story as just some unknown cat. variable, not (necessarily) text?
And have sth aabout text at a later time...