search

aqibsaeed / Multilabel-timeseries-classification-with-LSTM

Tensorflow implementation of paper: Learning to Diagnose with LSTM Recurrent Neural Networks.
Apache License 2.0
570 stars 189 forks source link

The format of the input data #3

Closed PerryDong closed 7 years ago

PerryDong commented 7 years ago

Can you offer the 'data.csv' used to train this model, as I want to know the format of the input data. Thank you!

aqibsaeed commented 7 years ago

The data set will look something like the figure below.

TEMP, PH etc. are all features, each red point represents feature value at one time point. Diagnosis vector (one-hot encoded) not shown in the picture will be class labels. Each training example will be a sequence of shape [1, time_steps, number_of_features] so your batch's shape will be [batch_size, time_steps, number_of_features].

Please see section on Dataset Description in paper for more detail. You can use subset of MIMIC-III. But as mentioned in the notebook pre-processing is required according to the use case.