Open Machine Learning Course

Author: Yury Kashnitsky, Data Scientist at Mail.ru Group
All content is distributed under the Creative Commons CC BY-NC-SA 4.0 license.

Assignment #6 (demo)

Exploring OLS, Lasso and Random Forest in a regression task

Fill in the missing code and choose answers in this web form.

In [1]:
import warnings
warnings.filterwarnings('ignore')
import numpy as np
import pandas as pd
from sklearn.metrics.regression import mean_squared_error
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import cross_val_score, train_test_split
from sklearn.linear_model import LinearRegression, LassoCV, Lasso
from sklearn.ensemble import RandomForestRegressor

We are working with UCI Wine quality dataset (no need to download it – it's already there, in course repo and in Kaggle Dataset).

In [2]:
data = pd.read_csv('../../data/winequality-white.csv', sep=';')
In [3]:
data.head()
Out[3]:
fixed acidity volatile acidity citric acid residual sugar chlorides free sulfur dioxide total sulfur dioxide density pH sulphates alcohol quality
0 7.0 0.27 0.36 20.7 0.045 45.0 170.0 1.0010 3.00 0.45 8.8 6
1 6.3 0.30 0.34 1.6 0.049 14.0 132.0 0.9940 3.30 0.49 9.5 6
2 8.1 0.28 0.40 6.9 0.050 30.0 97.0 0.9951 3.26 0.44 10.1 6
3 7.2 0.23 0.32 8.5 0.058 47.0 186.0 0.9956 3.19 0.40 9.9 6
4 7.2 0.23 0.32 8.5 0.058 47.0 186.0 0.9956 3.19 0.40 9.9 6
In [4]:
data.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4898 entries, 0 to 4897
Data columns (total 12 columns):
fixed acidity           4898 non-null float64
volatile acidity        4898 non-null float64
citric acid             4898 non-null float64
residual sugar          4898 non-null float64
chlorides               4898 non-null float64
free sulfur dioxide     4898 non-null float64
total sulfur dioxide    4898 non-null float64
density                 4898 non-null float64
pH                      4898 non-null float64
sulphates               4898 non-null float64
alcohol                 4898 non-null float64
quality                 4898 non-null int64
dtypes: float64(11), int64(1)
memory usage: 459.3 KB

Separate the target feature, split data in 7:3 proportion (30% form a holdout set, use random_state=17), and preprocess data with StandardScaler.

In [5]:
# y = None # you code here

# X_train, X_holdout, y_train, y_holdout = train_test_split # you code here
# scaler = StandardScaler()
# X_train_scaled = scaler.fit_transform # you code here
# X_holdout_scaled = scaler.transform # you code here

Linear regression

Train a simple linear regression model (Ordinary Least Squares).

In [6]:
# linreg = # you code here
# linreg.fit # you code here

Question 1: What are mean squared errors of model predictions on train and holdout sets?

In [7]:
# print("Mean squared error (train): %.3f" % # you code here
# print("Mean squared error (test): %.3f" % # you code here

Sort features by their influence on the target feature (wine quality). Beware that both large positive and large negative coefficients mean large influence on target. It's handy to use pandas.DataFrame here.

Question 2: Which feature this linear regression model treats as the most influential on wine quality?

In [8]:
# linreg_coef = pd.DataFrame # you code here
# linreg_coef.sort_values # you code here

Lasso regression

Train a LASSO model with $\alpha = 0.01$ (weak regularization). Again, set random_state=17.

In [9]:
# lasso1 = Lasso # you code here
# lasso1.fit # you code here

Which feature is the least informative in predicting wine quality, according to this LASSO model?

In [10]:
# lasso1_coef = pd.DataFrame # you code here
# lasso1_coef.sort_values # you code here

Train LassoCV with random_state=17 to choose the best value of $\alpha$ in 5-fold cross-validation.

In [11]:
# alphas = np.logspace(-6, 2, 200)
# lasso_cv = LassoCV # you code here
# lasso_cv.fit # you code here
In [12]:
# lasso_cv.alpha_

Question 3: Which feature is the least informative in predicting wine quality, according to the tuned LASSO model?

In [13]:
# lasso_cv_coef = pd.DataFrame # you code here
# lasso_cv_coef.sort_values # you code here

Question 4: What are mean squared errors of tuned LASSO predictions on train and holdout sets?

In [14]:
# print("Mean squared error (train): %.3f" % # you code here
# print("Mean squared error (test): %.3f" % # you code here

Random Forest

Train a Random Forest with out-of-the-box parameters, setting only random_state to be 17.

In [15]:
# forest = RandomForestRegressor # you code here
# forest.fit # you code here

Question 5: What are mean squared errors of RF model on the training set, in cross-validation (cross_val_score with scoring='neg_mean_squared_error' and other arguments left with default values) and on holdout set?

In [16]:
# print("Mean squared error (train): %.3f" % # you code here
# print("Mean squared error (cv): %.3f" % # you code here
# print("Mean squared error (test): %.3f" % # you code here

Tune the max_features and max_depth hyperparameters with GridSearchCV and again check mean cross-validation MSE and MSE on holdout set.

In [17]:
# forest_params = {'max_depth': list(range(10, 25)), 
#                  'min_samples_leaf': list(range(1, 8)),
#                  'max_features': list(range(6,12))}

# locally_best_forest = GridSearchCV # you code here
# locally_best_forest.fit # you code here
In [18]:
# locally_best_forest.best_params_, locally_best_forest.best_score_

Question 6: What are mean squared errors of tuned RF model in cross-validation (cross_val_score with scoring='neg_mean_squared_error' and other arguments left with default values) and on holdout set?

In [19]:
# print("Mean squared error (cv): %.3f" % # you code here
# print("Mean squared error (test): %.3f" % # you code here

Output RF's feature importance. Again, it's nice to present it as a DataFrame.
Question 7: What is the most important feature, according to the Random Forest model?

In [20]:
rf_importance = pd.DataFrame # you code here
rf_importance.sort_values # you code here
Out[20]:
<function pandas.core.frame.DataFrame.sort_values(self, by, axis=0, ascending=True, inplace=False, kind='quicksort', na_position='last')>

Make conclusions about the perdormance of the explored 3 models in this particular prediction task.