TIPS & TRICKS - How to Reshape Input Data for Long Short-Term Memory (LSTM) Networks in Tensorflow

This video is to provide guidance on how to convert your 1D or 2D data to the required 3D format of the LSTM input layer.

To make it easy to follow, you can download this notebook at Github and follow along with this step by step tutorial.

---------------------------------------------------------------------------------------------------------------------------

[jayktharwani]

To be honest, this video is among the best video I was looking for. I did not find any blog explaining this clearly. Really grateful to you.

[romankorotchenko]

Wow, I finally got a clear explanation and understood out how to organize LSTM for data from a set of sensors! Grand merci!

[pepe]

thank you very much for making this. this specific topic is very much overlooked by other tutorials.

[iftikhar]

Your videos help me a lot for understanding the input shape of data for sequenc model specially LSTM Model. Thanks again.

[mariaclaraassuncao]

Esse cara merece ser canonizado! melhor vídeo que encontrei

[kctay]

thank you for making this interesting video that's easy to understand. One question I have is what criteria you use to split samples by 48 ? I tried 36, 12 but both giving error.

[smithathurthi]

Thank you so much clear data prepartion for LSTM.

[batugkce]

Thx mister! Your work helped me a lot. I am appreciated to you.

[eighthseason]

Thank you! Is it possible to show how to scale training set and then use its scaler to scale test/validation set?

[alvinjamur]

this is wrong!! u cannot scale everything and then split data into train and test.

[vedantdalvi]

Thank you for the great explanation! Will the data processing be same for multiclass classification problem? I am doing classification of data with 5 classes

[lordyoun]

hello good tutorial .just want to know if it is still same if i want to do binary classification

[folashadeolaitan]

Thank you so much for the explanations. Is it necessary to use the np.stack()? If i have my X shape as (20050, 18), can't i just do np.reshape(X, 20050, 5, 18) assuming i want to use the last 5days data to predcit next day. I look forward to your response. Thank you.

[iftikhar]

when make the features Y you are wrong there. Y should ahead from the input features

[ahmedjamel]

Perfect explanation thank you. Just there is something I didn't get it. In y.append(df.iloc[i + 48, 6]) ; 6 refers to what ?