Develop is46 #48
Develop is46 #48
Conversation
…al error This push adds a flag to training that will allow you randomly generate the reverse complement examples for training in addition to the reference samples. The pyyaml req was added. The invalid interval bounds error was also patched by trying to get the input matrix and passing a matrix of zeros if the interval is not found.
| ### `--shuffle_cell_type` | ||
|
|
||
| If shuffle_cell_type, then shuffle training ROI cell type label. Default: `False` | ||
|
|
There was a problem hiding this comment.
Nice addition Tareain! This should make it easier to understand for the user.
| @@ -115,7 +116,8 @@ def run_training(args): | |||
| quant=args.quant, | |||
| batch_size=args.batch_size, | |||
| target_scale_factor=args.target_scale_factor, | |||
| shuffle_cell_type=args.shuffle_cell_type | |||
| shuffle_cell_type=args.shuffle_cell_type, | |||
| rev_comp_train=args.rev_comp | |||
| ) | |||
|
|
|||
| # Create keras.utils.sequence object from validation generator | |||
There was a problem hiding this comment.
Adding the rev_comp flag will help make our publication figures.
| # Our workaround for issue#42 is to provide a zero matrix for that position | ||
| try: | ||
| # Get the target matrix | ||
| target_vector = np.array(binding_stream.values(chrom_name, start, end)).T |
There was a problem hiding this comment.
I have a feeling that we do not need to Transpose this at the end.
| targets_batch.append(bin_vector) | ||
| except: | ||
| # TODO change length of array | ||
| target_vector = np.zeros(1024) |
There was a problem hiding this comment.
This should be the fix for the issue. Do we need to count the number of times this part is executed in the handful of TFs where it actually is implemented? (Might be overkill)
| @@ -24,24 +24,23 @@ def get_description(): | |||
| license="Apache-2.0", | |||
| include_package_data=True, | |||
| packages=find_packages(), | |||
| install_requires=["tensorflow==2.5.0",#add -gpu | |||
| install_requires=["tensorflow==2.5.0", | |||
There was a problem hiding this comment.
Should we be installing "tensorflow==2.5.0" and "tensorflow-gpu==2.5.0"? or should the user decide on his own?
| "deeplift", | ||
| "seaborn", | ||
| "pyyaml", |
| "deeplift", | ||
| "seaborn", | ||
| "pyyaml", | ||
| "graphviz", | ||
| "shap @ git+https://github.com/AvantiShri/shap.git@master#egg=shap", | ||
| "modisco @ git+https://github.com/XiaotingChen/tfmodisco.git@0.5.9.2#egg-modisco" |
There was a problem hiding this comment.
Can we ask Xiaoting to modify his install of modisco so we do not run into the sklearn issues when intalling maxatac? (detailed in Issue #47)
This pull request has made changes to the training data generator to fix the bug in #46. The fix that was implemented is here:
Additional updates to the docs were made. I also made changes to the
setup.pyfile by including the import of pyyaml. There is still an error with sklearn and the new updates that is fixed modifying the import statements inmodisco.