Make normalization of features standard practice in data processing

[khsjursen]

Add normalization of features to the data processing pipeline. Will likely not affect tree models, but needed for neural networks, so should make this standard for all models.

[khsjursen]

I tried training the Norway model with normalized features and it gave the same results as using unnormalized features with CustomXGBRegressor. I used MinMaxScaler from scikit-learn as shown below. Should still be implemented if we want to use other model architectures. Could be an option to have the user choose any scaler from scikit-learn based on their feature distributions. In any case we need to ensure that metadata is not scaled. My implementation was as follows:

# Normalize features using min-max scaling from scikit-learn
from sklearn.preprocessing import MinMaxScaler

# Get arrays of features+metadata and targets for training, df_train_X is training feature+metadata dataframe and df_train_y is target training dataframe
X_train_unnorm, y_train = df_train_X.values, df_train_y.values

# Initialize scaler
scaler = MinMaxScaler()

# Extract metadata columns such that these are not normalized
metadata_columns = X_train_unnorm[:, -3:]

# Extract remaining columns to be normalized
feature_columns = X_train_unnorm[:, :-3]

# Apply MinMaxScaler to feature columns
scaled_feature_columns = scaler.fit_transform(feature_columns)

# Combine scaled columns with metadata columns
X_train = np.hstack((scaled_feature_columns, metadata_columns))

# Get arrays of test features+metadata and test targets, df_test_X is test feature+metadata dataframe and df_test_y is target test dataframe
X_test_unnorm, y_test = df_test_X.values, df_test_y.values

# Extract metadata columns
metadata_columns_test = X_test_unnorm[:, -3:]

# Extract feature columns
feature_columns_test = X_test_unnorm[:, :-3]

# Apply MinMaxScaler fit to training features to the feature columns in test dataset
scaled_feature_columns_test = scaler.transform(feature_columns_test)

# Combine scaled columns with metadata columns
X_test = np.hstack((scaled_feature_columns_test, metadata_columns_test))