def plot_cumulative_features_importance(features,
threshold=0.90,
plot_size=(12, 8),
feature_type='features',
return_threshold=False):
"""
"""
plt.figure(figsize=plot_size)
# Number of features needed for threshold cumulative importance
importance_idx = np.min(np.where(features['cumulative_coefficient_frequency'] > threshold))
thr_percentage = 100 * threshold
required_features = importance_idx+1
l = '{} {} required for \n{:.0f}% cumulative importance.'.format(required_features,
feature_type,
thr_percentage)
# Cumulative importance plot
plt.plot(range(len(features)), features['cumulative_coefficient_frequency'], 'b-', label=l)
plt.xlabel('Number of {}'.format(feature_type.capitalize()), fontsize=12, labelpad=20)
plt.ylabel('Cumulative {} frequency'.format(feature_type.capitalize()), fontsize=12, labelpad=20)
plt.title('Cumulative {} Importance'.format(feature_type.capitalize()), fontsize=12, pad=20)
#plt.title(f'Cumulative Feature Importance\n\n{l}', fontsize=14, pad=20)
# Threshold vertical line plot
plt.vlines(importance_idx + 1, ymin=0, ymax=1.05, linestyles='--', colors='red')
plt.legend(loc='lower right', fontsize=10)
plt.tight_layout()
plt.show()
if return_threshold:
return required_features
def subplot_feature_importance(features, threshold=0.90, plot_size=(12, 8), return_data=False):
"""
"""
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=plot_size)
# First plot : Cumulative importance plot
# Number of features needed for threshold cumulative importance
importance_idx = np.min(np.where(features['cumulative_coefficient_frequency'] > threshold)) + 1
thr_percentage = 100 * threshold
# Legend label
l = '{} features required for \n{:.0f}% cumulative importance.'.format(importance_idx, thr_percentage)
ax1.plot(range(len(features)), features['cumulative_coefficient_frequency'], 'b-', label=l)
ax1.set_xlabel('Number of Features', fontsize=12, labelpad=20)
ax1.set_ylabel('Cumulative Coefficient frequency', fontsize=12, labelpad=20)
ax1.set_title('Cumulative Feature Importance', fontsize=14, pad=20)
#plt.title(f'Cumulative Feature Importance\n\n{l}', fontsize=14, pad=20)
# Threshold vertical line plot
ax1.vlines(importance_idx + 1, ymin=0, ymax=1.05, linestyles='--', colors='red')
ax1.legend(loc='lower right', fontsize=10)
# Second plot : n selected features
features.head(importance_idx).plot(x='feature', y='coefficient', kind='barh',
fontsize=12, figsize=plot_size, ax=ax2)
fig.gca().invert_yaxis()
model_label = features.index.name
ax2.set_title(f'{model_label} Top {importance_idx} Features', fontsize=14, pad=20)
ax2.set_xlabel('Coefficients', fontsize=12, labelpad=20)
ax2.set_ylabel('Features labels', fontsize=12, labelpad=20)
fig.tight_layout()
if return_data:
return features.head(importance_idx)
return fig
def plot_cumulative_features_importance(features, threshold=0.90, plot_size=(12, 8)):
"""
:param features: a dataframe which contains features data
:param threshold:
:param plot_size:
:return:
"""
plt.figure(figsize=plot_size)
# Number of features needed for threshold cumulative importance
importance_idx = np.min(np.where(features['cumulative_coefficient_frequency'] > threshold))
thr_percentage = 100 * threshold
l = '{} features required for {:.0f}% of cumulative importance.'.format(importance_idx+1, thr_percentage)
# Cumulative importance plot
plt.plot(range(len(features)), features['cumulative_coefficient_frequency'], 'b-', label=l)
plt.xlabel('Number of Features', fontsize=12, labelpad=20)
plt.ylabel('Cumulative Coefficient frequency', fontsize=12, labelpad=20)
plt.title('Cumulative Feature Importance', fontsize=14, pad=20)
# plt.title(f'Cumulative Feature Importance\n\n{l}', fontsize=14, pad=20)
# Threshold vertical line plot
plt.vlines(importance_idx + 1, ymin=0, ymax=1.05, linestyles='--', colors='red')
plt.legend(loc='lower right', fontsize=12)
plt.tight_layout()
plt.show()