#Find best k numeric features (Wrapper Method).

def plot_hists(dataset, nrows, ncols, title=""):
  original_cols = list(dataset.columns)
  plt.figure(figsize=(12, 16))
  for i in range(len(original_cols)):
    colname = original_cols[i]
    plt.subplot(nrows, ncols, i+1)
    plt.title(colname)
    dataset[colname].plot.hist(bins=200)

  plt.suptitle(title)
  plt.tight_layout()
  plt.subplots_adjust(top=0.95)
  plt.show()

def transform_data(dataset):
  original_cols = list(dataset.columns)
  power_trans = PowerTransformer(copy=True, method='yeo-johnson', standardize=True)
  transf_data = power_trans.fit_transform(dataset)
  df = pd.DataFrame(data=transf_data, columns=original_cols)
  return df

all_numeric_features = [
 'flesch_reading_ease',
 'smog_index',
 'flesch_kincaid_grade',
 'coleman_liau_index',
 'automated_readability_index',
 'dale_chall_readability_score',
 'difficult_words',
 'linsear_write_formula',
 'gunning_fog',
 'commentCount',
 'userHelpfulness',
 'body_length',
 'body_num_words',
 'num_spelling_mistakes',
 'total_helpful_votes_for_user',
 'total_votes_for_user',
 'reviewer_avg_rating',
 'subjectivity_avg',
 'subjectivity_std',
 'polarity_avg',
 'polarity_std',
]

train_set = data[data['review_date'] < train_max_date]
numeric_X_train = train_set[all_numeric_features].astype('f').reset_index(drop=True)
numeric_Y_train = train_set['helpful'].reset_index(drop=True)

# Transform data.
#plot_hists(numeric_X_train, 7, 3, "Original Distributions")
numeric_X_train = transform_data(numeric_X_train)
#plot_hists(numeric_X_train, 7, 3, "Modified Distributions")

sfs = SequentialFeatureSelector(RandomForestClassifier(n_estimators=20, max_depth=4),
                                k_features=10, forward=False, verbose=2, scoring='f1', cv=3)
sfs = sfs.fit(numeric_X_train.values, numeric_Y_train)
best_k_numeric = [all_numeric_features[feature_index] for feature_index in sfs.k_feature_idx_]

print('\nSequential Forward Selection Result:')
print(best_k_numeric)
print('CV Score:', sfs.k_score_)