cangermueller/deepcpg_legacy

Deep neural network for predicting methylation states of single cells

deepcpg_legacy's Issues

Couldn't load deepcpg

I ran the following under the repo directory:

python setup.py install

It shows that the preprocessing is successful without any errors. And when I do conda list, deepcpg also shows up. But I couldn't import it in python. Any idea?

Training example failed

I downloaded the example data with data.sh and tried to train a toy model with train.sh. Here is the error msg I got:

10000 training samples
10000 validation samples
INFO (2016-08-02 20:24:38,104): Train model
Epochs: 5
Samples: 10000
Batch size: 128
Learning rate: 0.000100
--------------------------------------------------------------------------------
Epoch 1/5
--------------------------------------------------------------------------------
Index: (0 - 10000)
Index: (0 - 10000)
Traceback (most recent call last):
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/theano/compile/function_module.py", line 866, in __call__
    self.fn() if output_subset is None else\
ValueError: dimension mismatch in args to gemm (128,7680)x(8000,5)->(128,5)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "./scripts/train.py", line 548, in <module>
    app.run(sys.argv)
  File "./scripts/train.py", line 133, in run
    return self.main(name, opts)
  File "./scripts/train.py", line 496, in main
    logger=lambda x: log.debug(x))
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/Keras-0.2.0-py3.5.egg/keras/models.py", line 998, in fit
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/Keras-0.2.0-py3.5.egg/keras/models.py", line 898, in _fit
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/theano/compile/function_module.py", line 879, in __call__
    storage_map=getattr(self.fn, 'storage_map', None))
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/theano/gof/link.py", line 325, in raise_with_op
    reraise(exc_type, exc_value, exc_trace)
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/six.py", line 685, in reraise
    raise value.with_traceback(tb)
  File "/cluster/zeng/code/research/software/miniconda/envs/py35/lib/python3.5/site-packages/theano/compile/function_module.py", line 866, in __call__
    self.fn() if output_subset is None else\
ValueError: dimension mismatch in args to gemm (128,7680)x(8000,5)->(128,5)
Apply node that caused the error: GpuDot22(GpuElemwise{mul,no_inplace}.0, <CudaNdarrayType(float32, matrix)>)
Toposort index: 75
Inputs types: [CudaNdarrayType(float32, matrix), CudaNdarrayType(float32, matrix)]
Inputs shapes: [(128, 7680), (8000, 5)]
Inputs strides: [(7680, 1), (5, 1)]
Inputs values: ['not shown', 'not shown']
Outputs clients: [[GpuElemwise{Add}[(0, 0)](GpuDot22.0, GpuDimShuffle{x,0}.0)]]

Any ideas?

Training on new data with sequence-only model

Hi Christof,

Great work! I have two questions:

1. Could you provide some guidelines on how to prepare the 'train.h5' files and so on for training on new data?
2. What should I do if I wish to train DeepCpG with the sequence module only (not using the nearby methylation level info) ? In the bioRxiv paper you seemed to have evaluated this.

Thanks!