Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting

Authors: Xingjian SHI, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, Wang-chun WOO

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that our Conv LSTM network captures spatiotemporal correlations better and consistently outperforms FC-LSTM and the state-of-theart operational ROVER algorithm for precipitation nowcasting.
Researcher Affiliation Collaboration Xingjian Shi Zhourong Chen Hao Wang Dit-Yan Yeung Department of Computer Science and Engineering Hong Kong University of Science and Technology {xshiab,zchenbb,hwangaz,dyyeung}@cse.ust.hk Wai-kin Wong Wang-chun Woo Hong Kong Observatory Hong Kong, China {wkwong,wcwoo}@hko.gov.hk
Pseudocode No The paper describes the model using mathematical equations and block diagrams (Fig. 2, Fig. 3) but does not provide structured pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper states 'Our implementations of the models are in Python with the help of Theano [3, 1]' and mentions a third-party open-source project for optical flow ('http://sourceforge.net/ projects/varflow/'), but it does not provide concrete access to the source code for the methodology described in this paper.
Open Datasets Yes 4.1 Moving-MNIST Dataset For this synthetic dataset, we use a generation process similar to that described in [21]... MNIST dataset: http://yann.lecun.com/exdb/mnist/
Dataset Splits Yes This generation process is repeated 15000 times, resulting in a dataset with 10000 training sequences, 2000 validation sequences, and 3000 testing sequences. ... Thus our radar echo dataset contains 8148 training sequences, 2037 testing sequences and 2037 validation sequences and all the sequences are 20 frames long (5 for the input and 15 for the prediction).
Hardware Specification Yes We run all the experiments on a computer with a single NVIDIA K20 GPU.
Software Dependencies No The paper states 'Our implementations of the models are in Python with the help of Theano [3, 1]' and mentions a 'MATLAB function fspecial( disk , 10)' and an 'open-source project to calculate the optical flow: http://sourceforge.net/ projects/varflow/', but it does not specify version numbers for Python or Theano, which are key software components.
Experiment Setup Yes We train all the LSTM models by minimizing the cross-entropy loss4 using back-propagation through time (BPTT) [2] and RMSProp [24] with a learning rate of 10 3 and a decay rate of 0.9. Also, we perform early-stopping on the validation set. ... For our Conv LSTM network, we set the patch size to 4 4 so that each 64 64 frame is represented by a 16 16 16 tensor. ... All the input-to-state and state-to-state kernels are of size 5 5. ... We set the patch size to 2 and train a 2-layer Conv LSTM network with each layer containing 64 hidden states and 3 3 kernels.