C5W2 Emoji_v3a LSTM Error 'expected ndim=3, found ndim=2'

Course Q&A Deep Learning Specialization Sequence Models
1

I’ve been stuck on this for a few hours. I feel I’m missing something obvious.

When I run the unit test for my attempt at implementing Emojify_V2 I get the following error:

ValueError: Input 0 of layer lstm_25 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [4, 2]

If I print(embeddings.shape) that returns (4, 2).
I’m guessing I either shouldn’t be passing embeddings into my LSTM object, or I need to expand its dimensions? Can anyone suggest which direction to go in please?

2

“embeddings” is the data that you pass to the LSTM layer. It should be set off in its own set of parenthesis, after you’ve created the LSTM layer, by using X = LSTM(…)(embeddings)

3

Thanks. Yes, sorry I think I’m doing that. I’m not sure how much code I’m allowed to post here, but I have:
X = LSTM(128, return_sequences=True)(embeddings)

4

Maybe there is an issue in your pretrained_embedding_layer().

5

Or maybe you don’t have the correct arguments for the Input(…) layer that creates “sentence_indices”.

6

Thanks for your help. I’ve got it working now. In my case my problem was well before the line with the error. I had implemented the following line incorrectly:
sentence_indices = ...
and I guess it takes a few lines to blow up.