scalar values from trained neuralhydrology model can be updated to avoid pickle.

[jmframe]

Neural Hydrology has updated the way they save the scalars necessary to process input and output data. Training any new LSTM models with neural hydrology would require reading in scalar values from yaml format, rather than pickle files. This is good, because it means that the BMI LSTM is no longer dependent on pickle, at all. Screenshot below shows the changes needed to read in scalars from yaml format. Also pasted below is updated code to read in scalars from yaml. Finally, this change can also be found in this PR from my branch:

jmframe#11

` scaler_file = os.path.join(self.cfg_train['run_dir'], 'train_data', 'train_data_scaler.yml')

    with open(scaler_file, 'r') as f:
        scaler_data = yaml.safe_load(f)

    self.train_data_scaler = scaler_data

    self.attribute_means = scaler_data.get('attribute_means', {})
    self.attribute_stds = scaler_data.get('attribute_stds', {})
    self.feature_scale = {k: v['data'] for k, v in scaler_data['xarray_feature_scale']['data_vars'].items()}
    self.feature_center = {k: v['data'] for k, v in scaler_data['xarray_feature_center']['data_vars'].items()}
    if self.verbose > 1:
        print(self.feature_center)
        print(self.feature_scale)
        print(self.attribute_means)
        print(self.attribute_stds)

#------------------------------------------------------------`

"""Mean and standard deviation for the inputs and LSTM outputs""" self.out_mean = self.train_data_scaler['xarray_feature_center']['data_vars'][self.cfg_train['target_variables'][0]]['data'] self.out_std = self.train_data_scaler['xarray_feature_scale']['data_vars'][self.cfg_train['target_variables'][0]]['data'] self.input_mean.extend([self.train_data_scaler['xarray_feature_center']['data_vars'][x]['data'] for x in self.cfg_train['dynamic_inputs']]) self.input_std.extend([self.train_data_scaler['xarray_feature_scale']['data_vars'][x]['data'] for x in self.cfg_train['dynamic_inputs']]) self.input_std.extend([self.train_data_scaler['attribute_stds'][x] for x in self.cfg_train['static_attributes']]) self.input_std = np.array(self.input_std)