import neralbinder.neuralbindhelpers.plot_helper as plot_helper from neuralbinder.deepomics import neuralnetwork as nn from neuralbinder.deepomics import utils, fit #--------------------------------------------------------------------------------------- models = ['clip_conv_net', 'clip_residualbind'] ss_types = ['seq', 'pu'] window = 200 # dataset path dataset_path = '/media/peter/storage/encode_eclip/eclip_datasets' # set results path results_path = helper.make_directory('../../results', 'encode_eclip') saliency_results_path = utils.make_directory(results_path, 'saliency_ensemble') # get list of .h5 files in dataset path file_names = helper.get_file_names(dataset_path) # loop through models for model in models: # model results path model_path = helper.make_directory(results_path, model) saliency_model_path = helper.make_directory(saliency_results_path, model) # loop through secondary structure types for ss_type in ss_types:
from neuralbinder.deepomics import neuralnetwork as nn from neuralbinder.deepomics import utils, fit, metrics, saliency, visualize, init #--------------------------------------------------------------------------------------- num_saliencies = [200] num_filters = 5 # different deep learning models to try out models = ['affinity_conv_net', 'affinity_residualbind', 'affinity_all_conv_net'] normalize_method = 'log_norm' # 'clip_norm' ss_types = ['seq', 'pu'] data_path = neuralbinder.get_datasets('rnacompete2009.h5') trained_path = '../../results/RNAcompete_2009' results_path = utils.make_directory(trained_path, 'motifs') #--------------------------------------------------------------------------------------- # classifier model def classifier_model(input_shape, output_shape, num_filters): # create model layer1 = {'layer': 'input', 'input_shape': input_shape } layer2 = {'layer': 'conv1d', 'num_filters': num_filters, 'filter_size': 11, 'activation': 'sigmoid', 'W': init.HeUniform(),
from neuralbinder.deepomics import neuralnetwork as nn from neuralbinder.deepomics import utils, fit #--------------------------------------------------------------------------------------- num_epochs = 100 batch_size = 100 # different deep learning models to try out models = ['affinity_conv_net', 'affinity_residualbind', 'affinity_all_conv_net'] normalize_method = 'log_norm' # 'clip_norm' ss_types = ['seq', 'pu'] data_path = neuralbinder.get_datasets('rnacompete2013.h5') trained_path = '../../results/RNAcompete_2013' results_path = utils.make_directory(trained_path, 'saliency') #--------------------------------------------------------------------------------------- # get list of rnacompete experiments experiments = helper.get_experiments_hdf5(data_path) # loop over different secondary structure contexts for ss_type in ss_types: print('input data: ' + ss_type) sstype_path = os.path.join(trained_path, normalize_method+'_'+ss_type) # directory to save parameters of saliency saliency_sstype_path = helper.make_directory(results_path, normalize_method+'_'+ss_type) saliency_sstype_path = helper.make_directory(saliency_sstype_path, 'ensemble')