lc_file=base_out_folder + "lcs/lc_fold_" + str(cur_fold))
ids, projection, classes = model.get_projection(training_datasets)
cross_validation.save_projections(projection,
base_out_folder + "/projections/",
ids)
"""
Main method.
Check the run.sh script for the list of parameters.
"""
if __name__ == "__main__":
params_sys = eval(sys.argv[1])
if params_sys["base_model"] == "module":
dl = load_dataset.DatasetLoader()
params = {"epochs": params_sys["epochs"]}
dataset = dl.load_dataset(
params_sys["feature_type"],
binary_features=params_sys["binary_feature"] == "True")
run_base_models(params_sys["base_folder"], params, params_sys["seed"],
params_sys["fold"], dataset)
if params_sys["base_model"] == "joined_seq":
create_dir_structure(params_sys["base_folder"])
selected_modules = []
available_modules = list(params_sys["models_to_load"])
print("Available models:", str(available_modules))
has_improved = True
def add_loss_op(self, pred):
loss = utils.cross_entropy(pred, self.y_)
return loss
def add_training_op(self, loss):
train_op = utils.adam_opt(loss, self.config.lr, self.loader.num_trainbatches(), 0.98)
return train_op
if len(argv) > 1:
chromosome = int(argv[1])
else:
chromosome=21
config = Config()
loader = load_dataset.DatasetLoader(chromosome=chromosome, windowSize=config.window, #custom_load_dataset.DatasetLoader(chromosome=chromosome, windowSize=config.window,
testBatchSize=config.test_batch_size,
seed=1, test_frac=0.025, pos_frac=0.5, load_coverage=False, load_entire=True)
'''for i in range(100000):
loader.load_chromosome_window_batch(10, 100)
print(loader.load_chromosome_window_batch(10, 5))
print(loader.chrom_index)
exit()'''
#print(loader.test_data)
conv_net = SimpleConv(config, loader)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
train_op = utils.adam_opt(loss, self.config.lr,
self.loader.num_trainbatches(), 0.98)
return train_op
if len(argv) < 2:
window = 50
else:
window = int(argv[1])
print("window {}".format(window))
config = Config(window)
loader = load_dataset.DatasetLoader(chromosome=21,
windowSize=config.window,
batchSize=config.batch_size,
testBatchSize=config.test_batch_size,
seed=1,
test_frac=0.025,
pos_frac=0.5,
load_coverage=False)
#loader = load_full_dataset_sample_one_pos.DatasetLoader(windowSize=config.window, batchSize=config.batch_size, testBatchSize=config.test_batch_size, seed=1, test_frac=0.05, load_coverage=False)
conv_net = SimpleConv(config, loader, plotTrain=True)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
#losses, val_accuracies = conv_net.fit(sess, save=True)
#conv_net.predictAll(sess, save=True)
#all_results = conv_net.hard_examples(sess)
#hard_positives = [x for x in all_results if x[1]]
def run_joined_models_tree(base_out_folder, base_in_folder, models_to_load,
cur_fold, s):
create_dir_structure(base_out_folder)
# set random seeds
set_seeds(s)
# checks if result file already exists. if so, do not run the algorithm.
if cur_fold != -1:
f_to_check = base_out_folder + "/predictions/predictions_fold_" + str(
cur_fold)
if file_exists(f_to_check):
print("not running")
print("file exists", f_to_check)
return
# load the testing and training entrez ids for this fold
training_entrez, testing_entrez = read_entrez_indexes(
base_in_folder + models_to_load[0] + "/folds/fold_" +
str(cur_fold))
testing_predictions = [
base_in_folder + model_name + "/predictions/predictions_fold_" +
str(cur_fold) for model_name in models_to_load
]
testing_dataset = merge_predictions(testing_predictions,
models_to_load)
training_predictions = [
base_in_folder + model_name +
"/predictions_train/predictions_fold_" + str(cur_fold)
for model_name in models_to_load
]
training_dataset = merge_predictions(training_predictions,
models_to_load)
al = load_dataset.DatasetLoader() ####
classes = pandas.read_csv(al.base_path + "class_labels.csv",
sep=",",
header=0,
na_values=["?"])
classes.dropna(subset=["class_Brain.Alzheimer"], inplace=True)
classes.columns = [
"".join(c if c.isalnum() else "_" for c in str(x))
for x in classes.columns
] ########################################################
class_indexes = classes.columns[1:]
classes.set_index("entrezId", inplace=True)
training_dataset = training_dataset.join(classes)
training_dataset.class_indexes = class_indexes
testing_dataset = testing_dataset.join(classes)
testing_dataset.class_indexes = class_indexes
model = train_model.BoostedTreeModel()
model.train(training_dataset, testing_dataset)
ids, predictions, classes = model.evaluate(testing_dataset)
CrossValidation.write_results(ids, predictions, classes, cur_fold,
base_out_folder + "/predictions/")
else:
if file_exists(base_out_folder + "/projections/_projection.csv"):
print("not running")
return
training_projections = [
base_in_folder + model_name +
"/projections/fold_train_-1_projection.csv"
for model_name in models_to_load
]
training_datasets = merge_projections(training_projections, None)
model = train_model.JoinedModel(epochs)
model.train(training_datasets,
training_datasets,
lc_file=base_out_folder + "lcs/lc_fold_" + str(cur_fold))
ids, projection, classes = model.get_projection(training_datasets)
cross_validation.save_projections(projection,
base_out_folder + "/projections/",
ids)