Various flavours of RNNs implemented in Tensorflow. These models use the new Estimator and Dataset APIs, and therefore can be used out of the box for distributed training.
The full dataset is ~500MB and can be downloaded at : https://www.kaggle.com/c/web-traffic-time-series-forecasting/data
data_preprocessing.py : script to pre-process the data. Further data preparation (such as one-hot encoding) takes place within tensorflow.
gru_contextual.py : RNN with GRU cell and contextual features
rnn_baseline.py : Baseline RNN with no contextual features and simple cell
rnn_contextual.py : Baseline RNN + contextual features
seq2seq.py : Encoder Decoder architecture
utils.py : contains functions used by all models, including preparation of input pipeline for tensorflow. Also includes a custom new metric written to be used with EvalSpec : SMAPE.
parse_input_parameters.py : utility script to parse command line args