can't concat when embedding_size is set to "auto"

[dev-wei]

Describe the bug(问题描述) When set the embedding size to "auto", the Concatenate layer can't merge all input Embedding with different size at axis=2

def concat_fun(inputs, axis=-1): if len(inputs) == 1: return inputs[0] else: return Concatenate(axis=axis)(inputs)

To Reproduce(复现步骤) Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. ValueError: A Concatenate layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 1, 36), (None, 1, 30), (None, 1, 6), (None, 1, 12), (None, 1, 12), (None, 1, 30), (None, 1, 12)]

Operating environment(运行环境):

• python version [e.g. 3.4, 3.6]
• tensorflow version [e.g. 1.4.0, 1.12.0]
• deepctr version [e.g. 0.2.3,]

Additional context Add any other context about the problem here.

[shenweichen]

can you concat the emb list with axis=-1?

[dev-wei]

@shenweichen thanks for the quick response.

do you mean change "fm_input = concat_fun(deep_emb_list, axis=1)" to "fm_input = concat_fun(deep_emb_list, axis=-1)"

?

[shenweichen]

the auto embedding mode is not supported when you use fm based model.

[dev-wei]

that sounds clear. Perhaps, we should add a validation error out there.

Also, I want to gain your opinion on "what is the right embedding size, when dataset has a number of sparse fields".

[shenweichen]

Thanks for your advice

Usually I search for a size in 4 8 12 16 32 64 etc

[dev-wei]

Thanks for the sharing. I will do extra tuning over there.

Currently, I am evaluate movie-len 100k with LR, the mse I got, based on epoch=10, embedding size=8 is not amazing.

the train mse is 0.7317, and test mse is 0.9476 - for nfm model.

Do you have some metrics you could share?

I'd continue do more.

[dev-wei]

btw, I love this repo and your quickness on responding. Good work.

[shenweichen]

Sorry I don’t have any metrics on movie-len ~