Hello, I'm Diego. I'm training NNs using ClearML. I've had some problems when cloning experiments and changing hyper params. My train script loads  params  from a json file and then connects them to the task. The code has the following structure:

` # imports
(...)
with open(args.params) as fid:
params = json.load(fid)

some params.setdefault()

params.setdefault('exp','My Training')
params.setdefault('batch_size', 32)
(...)

task_name = params['exp']

dataset = Dataset.get(dataset_name=params['dataset'], dataset_project='My Datasets')
dataset_dir = dataset.get_local_copy()
df_train = pd.read_pickle(op.join(dataset_dir, 'df_train.pkl'))
df_test = pd.read_pickle(op.join(dataset_dir, 'df_val.pkl'))

task = Task.init(task_name=task_name,
project_name='My Project',
reuse_last_task_id=False,
auto_connect_streams=False)

task.connect(params)
train_dnn(df_train, df_test, params) `
The problem is that when I clone an experiment and change the hyper params some change and some remain the same as the original ones. I have also deleted some hyper-params but they appear again when training starts.