Data type Error while running basset_train.lua #25
Comments
|
As far as you can tell, did the torch-hdf5 library install properly? Can you start torch and import it? Alternatively, what version of HDF5 do you have installed? I noticed awhile back that the newest version caused a few things to break. |
|
Thanks Dave. My whole new-bie exposure to lua and torch is through Basset. I hope I am not super messy here:
ii hdf5-helpers 1.8.16+docs-4ubuntu1 amd64 Hierarchical Data Format 5 (HDF5) - Helper tools sam@ubuntu-supermicro:~/softwares/basset/data/dnase_bed_files$ th ______ __ | Torch7 th> require 'hdf5' |
|
Hmm OK that all looks fine. Could you send me your HDF5 file to test on my system? Maybe the files that you used to make it, too, if that's not too challenging. |
|
I have sent a link to the files uploaded on google-drive to your email id (mentioned in the basset paper). |
|
It seems to be working fine for me. Maybe try re-installing the torch-hdf5 package; you must have the wrong one somehow. Run the following commands in the "src" directory":
And try running again. |
|
Dave, I did what you suggested! And it's working! Thanks a lot again! |
|
Great to hear. I had to modify DeepMind's version of the torch-hdf5 code a bit to enable 2 byte floats, so you must've installed their version instead of mine. |
|
yeah, spot on! I saw that somehow both theirs and yours were co-existing and there was a conflict. Got rid of both first, and then reinstalled yours one. |
|
I also run the your suggested commands in the "src" directory" . It did success to solve the problem 'Cannot support reading float data with size = 2 bytes', Could you please suggest something? |
|
Yea HDF 1.10 breaks the DeepMind library. I use HDF 1.8 |
Hi Dave,
I followed the instructions mentioned in the tutorial file: new_data_iso.ipynb
Then ran this:
$ basset_train.lua -job params.txt -save mg_dnase_cnn learn_mg_dnase.h5
(previously, learn_mg_dnase.bed, learn_mg_dnase_act.txt, learn_mg_dnase.fa and learn_mg_dnase.h5 were created as per the instructions with your scripts)
I get this error message : "ffi.lua:332: Cannot support reading float data with size = 2 bytes"
Question: Is this error related to the datatype in the .h5 file?
Could you please suggest something?
Thanks!
(Please see the error details below)
{}
nn.Sequential {
input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> (9) -> output: nn.SpatialConvolution(4 -> 10, 10x1)
(2): nn.SpatialBatchNormalization
(3): nn.ReLU
(4): nn.Reshape(5910)
(5): nn.Linear(5910 -> 500)
(6): nn.BatchNormalization
(7): nn.ReLU
(8): nn.Linear(500 -> 9)
(9): nn.Sigmoid
}
/home/sam/softwares/torch/install/bin/luajit: ...e/sam/softwares/torch/install/share/lua/5.1/hdf5/ffi.lua:332: Cannot support reading float data with size = 2 bytes
stack traceback:
[C]: in function 'error'
...e/sam/softwares/torch/install/share/lua/5.1/hdf5/ffi.lua:332: in function '_getTorchType'
...m/softwares/torch/install/share/lua/5.1/hdf5/dataset.lua:88: in function 'getTensorFactory'
...m/softwares/torch/install/share/lua/5.1/hdf5/dataset.lua:138: in function 'partial'
/home/sam/softwares/basset/src/batcher.lua:31: in function 'next'
/home/sam/softwares/basset/src/convnet.lua:884: in function 'train_epoch'
/home/sam/softwares/basset/src/basset_train.lua:148: in main chunk
[C]: in function 'dofile'
...ares/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00405d50
Epoch # 1
The text was updated successfully, but these errors were encountered: