Caffe for time series Data

[Abhishek Sharma]

I am working on 1D data to classify it into two classes. I am using caffe. My data has 2000 samples. I have made an hdf5 file containing two data sets - 'data and label'.  I am using the network definition similar to MNIST but with hdf5 input. But, every time I try to train the network it gives the following error:

I0629 14:53:52.153496  3751 layer_factory.hpp:74] Creating layer mnist

I0629 14:53:52.153542  3751 net.cpp:90] Creating Layer mnist

I0629 14:53:52.153568  3751 net.cpp:368] mnist -> data

I0629 14:53:52.153614  3751 net.cpp:368] mnist -> label

I0629 14:53:52.153650  3751 net.cpp:120] Setting up mnist

I0629 14:53:52.153671  3751 hdf5_data_layer.cpp:80] Loading list of HDF5 filenames from: examples/1D_hdf5/Data/Train.txt

I0629 14:53:52.153738  3751 hdf5_data_layer.cpp:94] Number of HDF5 files: 1

I0629 14:53:52.193681  3751 net.cpp:127] Top shape: 47 1 1 2000 (94000)

I0629 14:53:52.193735  3751 net.cpp:127] Top shape: 47 1 (47)

I0629 14:53:52.193758  3751 layer_factory.hpp:74] Creating layer conv1

I0629 14:53:52.193792  3751 net.cpp:90] Creating Layer conv1

I0629 14:53:52.193814  3751 net.cpp:410] conv1 <- data

I0629 14:53:52.193840  3751 net.cpp:368] conv1 -> conv1

I0629 14:53:52.193871  3751 net.cpp:120] Setting up conv1

F0629 14:53:52.194491  3751 blob.cpp:28] Check failed: shape[i] >= 0 (-3 vs. 0) 

*** Check failure stack trace: ***

    @     0x7f4b774e5daa  (unknown)

    @     0x7f4b774e5ce4  (unknown)

    @     0x7f4b774e56e6  (unknown)

    @     0x7f4b774e8687  (unknown)

    @     0x7f4b7786c2e7  caffe::Blob<>::Reshape()

    @     0x7f4b7786c4c9  caffe::Blob<>::Reshape()

    @     0x7f4b778d22a8  caffe::BaseConvolutionLayer<>::Reshape()

    @     0x7f4b7785b342  caffe::Net<>::Init()

    @     0x7f4b7785d0a2  caffe::Net<>::Net()

    @     0x7f4b77843ae0  caffe::Solver<>::InitTrainNet()

    @     0x7f4b77844a83  caffe::Solver<>::Init()

    @     0x7f4b77844c56  caffe::Solver<>::Solver()

    @           0x40c2f0  caffe::GetSolver<>()

    @           0x4062c3  train()

    @           0x4048d1  main

    @     0x7f4b769f7ec5  (unknown)

    @           0x404e7d  (unknown)

    @              (nil)  (unknown)

Aborted (core dumped)

Pls help!!

[Abhishek Sharma]

This is the train.h5 file.

HDF5 train.h5 

Group '/' 

    Dataset 'data' 

        Size:  2000x1x1x2399

        MaxSize:  2000x1x1x2399

        Datatype:   H5T_IEEE_F64LE (double)

        ChunkSize:  []

        Filters:  none

        FillValue:  0.000000

    Dataset 'label' 

        Size:  1x2399

        MaxSize:  1x2399

        Datatype:   H5T_IEEE_F64LE (double)

        ChunkSize:  []

        Filters:  none

        FillValue:  0.000000

[Leif Blaese]

I0629 14:53:52.193758  3751 layer_factory.hpp:74] Creating layer conv1

I0629 14:53:52.193792  3751 net.cpp:90] Creating Layer conv1

I0629 14:53:52.193814  3751 net.cpp:410] conv1 <- data

I0629 14:53:52.193840  3751 net.cpp:368] conv1 -> conv1

I0629 14:53:52.193871  3751 net.cpp:120] Setting up conv1

F0629 14:53:52.194491  3751 blob.cpp:28] Check failed: shape[i] >= 0 (-3 vs. 0) 

 

Well, caffe complains about the shape of the data. It seems like one dimension of the kernel is 0? Maybe the kernel has a larger size (in the height dimension) than your input. Change the input to a 2-D format or change the kernel so that is has a kernel_h of 1 (look up convolutional parameters in the caffe tutorial). Try and play around with the parameters until they fit... ;)

For further help, show the whole architecture and not only the first two layers so we know where the error is.