02_end_to_end_machine_learning_project-checkpoint.ipynb @master — view markup · raw · history · blame
Chapter 2 – End-to-end Machine Learning project
Welcome to Machine Learning Housing Corp.! Your task is to predict median house values in Californian districts, given a number of features from these districts.
This notebook contains all the sample code and solutions to the exercices in chapter 2.
Note: You may find little differences between the code outputs in the book and in these Jupyter notebooks: these slight differences are mostly due to the random nature of many training algorithms: although I have tried to make these notebooks' outputs as constant as possible, it is impossible to guarantee that they will produce the exact same output on every platform. Also, some data structures (such as dictionaries) do not preserve the item order. Finally, I fixed a few minor bugs (I added notes next to the concerned cells) which lead to slightly different results, without changing the ideas presented in the book.
Setup¶
First, let's make sure this notebook works well in both python 2 and 3, import a few common modules, ensure MatplotLib plots figures inline and prepare a function to save the figures:
# To support both python 2 and python 3
from __future__ import division, print_function, unicode_literals
# Common imports
import numpy as np
import os
# to make this notebook's output stable across runs
np.random.seed(42)
# To plot pretty figures
%matplotlib inline
import matplotlib
import matplotlib.pyplot as plt
plt.rcParams['axes.labelsize'] = 14
plt.rcParams['xtick.labelsize'] = 12
plt.rcParams['ytick.labelsize'] = 12
# Where to save the figures
PROJECT_ROOT_DIR = "."
CHAPTER_ID = "end_to_end_project"
IMAGES_PATH = os.path.join(PROJECT_ROOT_DIR, "images", CHAPTER_ID)
def save_fig(fig_id, tight_layout=True, fig_extension="png", resolution=300):
path = os.path.join(IMAGES_PATH, fig_id + "." + fig_extension)
print("Saving figure", fig_id)
if tight_layout:
plt.tight_layout()
plt.savefig(path, format=fig_extension, dpi=resolution)
# Ignore useless warnings (see SciPy issue #5998)
import warnings
warnings.filterwarnings(action="ignore", message="^internal gelsd")
Get the data¶
import os
import tarfile
from six.moves import urllib
DOWNLOAD_ROOT = "https://raw.githubusercontent.com/ageron/handson-ml/master/"
HOUSING_PATH = os.path.join("datasets", "housing")
HOUSING_URL = DOWNLOAD_ROOT + "datasets/housing/housing.tgz"
def fetch_housing_data(housing_url=HOUSING_URL, housing_path=HOUSING_PATH):
if not os.path.isdir(housing_path):
os.makedirs(housing_path)
tgz_path = os.path.join(housing_path, "housing.tgz")
urllib.request.urlretrieve(housing_url, tgz_path)
housing_tgz = tarfile.open(tgz_path)
housing_tgz.extractall(path=housing_path)
housing_tgz.close()
fetch_housing_data()
import pandas as pd
def load_housing_data(housing_path=HOUSING_PATH):
csv_path = os.path.join(housing_path, "housing.csv")
return pd.read_csv(csv_path)
housing = load_housing_data()
housing.head()
housing.info()
housing["ocean_proximity"].value_counts()
housing.describe()
%matplotlib inline
import matplotlib.pyplot as plt
housing.hist(bins=50, figsize=(20,15))
save_fig("attribute_histogram_plots")
plt.show()
# to make this notebook's output identical at every run
np.random.seed(42)
import numpy as np
# For illustration only. Sklearn has train_test_split()
def split_train_test(data, test_ratio):
shuffled_indices = np.random.permutation(len(data))
test_set_size = int(len(data) * test_ratio)
test_indices = shuffled_indices[:test_set_size]
train_indices = shuffled_indices[test_set_size:]
return data.iloc[train_indices], data.iloc[test_indices]
train_set, test_set = split_train_test(housing, 0.2)
print(len(train_set), "train +", len(test_set), "test")
from zlib import crc32
def test_set_check(identifier, test_ratio):
return crc32(np.int64(identifier)) & 0xffffffff < test_ratio * 2**32
def split_train_test_by_id(data, test_ratio, id_column):
ids = data[id_column]
in_test_set = ids.apply(lambda id_: test_set_check(id_, test_ratio))
return data.loc[~in_test_set], data.loc[in_test_set]
The implementation of test_set_check() above works fine in both Python 2 and Python 3. In earlier releases, the following implementation was proposed, which supported any hash function, but was much slower and did not support Python 2:
import hashlib
def test_set_check(identifier, test_ratio, hash=hashlib.md5):
return hash(np.int64(identifier)).digest()[-1] < 256 * test_ratio
If you want an implementation that supports any hash function and is compatible with both Python 2 and Python 3, here is one:
def test_set_check(identifier, test_ratio, hash=hashlib.md5):
return bytearray(hash(np.int64(identifier)).digest())[-1] < 256 * test_ratio
housing_with_id = housing.reset_index() # adds an `index` column
train_set, test_set = split_train_test_by_id(housing_with_id, 0.2, "index")
housing_with_id["id"] = housing["longitude"] * 1000 + housing["latitude"]
train_set, test_set = split_train_test_by_id(housing_with_id, 0.2, "id")
test_set.head()
from sklearn.model_selection import train_test_split
train_set, test_set = train_test_split(housing, test_size=0.2, random_state=42)
test_set.head()
housing["median_income"].hist()
# Divide by 1.5 to limit the number of income categories
housing["income_cat"] = np.ceil(housing["median_income"] / 1.5)
# Label those above 5 as 5
housing["income_cat"].where(housing["income_cat"] < 5, 5.0, inplace=True)
housing["income_cat"].value_counts()
housing["income_cat"].hist()
from sklearn.model_selection import StratifiedShuffleSplit
split = StratifiedShuffleSplit(n_splits=1, test_size=0.2, random_state=42)
for train_index, test_index in split.split(housing, housing["income_cat"]):
strat_train_set = housing.loc[train_index]
strat_test_set = housing.loc[test_index]
strat_test_set["income_cat"].value_counts() / len(strat_test_set)
housing["income_cat"].value_counts() / len(housing)
def income_cat_proportions(data):
return data["income_cat"].value_counts() / len(data)
train_set, test_set = train_test_split(housing, test_size=0.2, random_state=42)
compare_props = pd.DataFrame({
"Overall": income_cat_proportions(housing),
"Stratified": income_cat_proportions(strat_test_set),
"Random": income_cat_proportions(test_set),
}).sort_index()
compare_props["Rand. %error"] = 100 * compare_props["Random"] / compare_props["Overall"] - 100
compare_props["Strat. %error"] = 100 * compare_props["Stratified"] / compare_props["Overall"] - 100
compare_props
for set_ in (strat_train_set, strat_test_set):
set_.drop("income_cat", axis=1, inplace=True)
Discover and visualize the data to gain insights¶
housing = strat_train_set.copy()
housing.plot(kind="scatter", x="longitude", y="latitude")
save_fig("bad_visualization_plot")
housing.plot(kind="scatter", x="longitude", y="latitude", alpha=0.1)
save_fig("better_visualization_plot")
The argument sharex=False fixes a display bug (the x-axis values and legend were not displayed). This is a temporary fix (see: https://github.com/pandas-dev/pandas/issues/10611). Thanks to Wilmer Arellano for pointing it out.
housing.plot(kind="scatter", x="longitude", y="latitude", alpha=0.4,
s=housing["population"]/100, label="population", figsize=(10,7),
c="median_house_value", cmap=plt.get_cmap("jet"), colorbar=True,
sharex=False)
plt.legend()
save_fig("housing_prices_scatterplot")
import matplotlib.image as mpimg
california_img=mpimg.imread(PROJECT_ROOT_DIR + '/images/end_to_end_project/california.png')
ax = housing.plot(kind="scatter", x="longitude", y="latitude", figsize=(10,7),
s=housing['population']/100, label="Population",
c="median_house_value", cmap=plt.get_cmap("jet"),
colorbar=False, alpha=0.4,
)
plt.imshow(california_img, extent=[-124.55, -113.80, 32.45, 42.05], alpha=0.5,
cmap=plt.get_cmap("jet"))
plt.ylabel("Latitude", fontsize=14)
plt.xlabel("Longitude", fontsize=14)
prices = housing["median_house_value"]
tick_values = np.linspace(prices.min(), prices.max(), 11)
cbar = plt.colorbar()
cbar.ax.set_yticklabels(["$%dk"%(round(v/1000)) for v in tick_values], fontsize=14)
cbar.set_label('Median House Value', fontsize=16)
plt.legend(fontsize=16)
save_fig("california_housing_prices_plot")
plt.show()
corr_matrix = housing.corr()
corr_matrix["median_house_value"].sort_values(ascending=False)
# from pandas.tools.plotting import scatter_matrix # For older versions of Pandas
from pandas.plotting import scatter_matrix
attributes = ["median_house_value", "median_income", "total_rooms",
"housing_median_age"]
scatter_matrix(housing[attributes], figsize=(12, 8))
save_fig("scatter_matrix_plot")
housing.plot(kind="scatter", x="median_income", y="median_house_value",
alpha=0.1)
plt.axis([0, 16, 0, 550000])
save_fig("income_vs_house_value_scatterplot")
housing["rooms_per_household"] = housing["total_rooms"]/housing["households"]
housing["bedrooms_per_room"] = housing["total_bedrooms"]/housing["total_rooms"]
housing["population_per_household"]=housing["population"]/housing["households"]
Note: there was a bug in the previous cell, in the definition of the rooms_per_household attribute. This explains why the correlation value below differs slightly from the value in the book (unless you are reading the latest version).
corr_matrix = housing.corr()
corr_matrix["median_house_value"].sort_values(ascending=False)
housing.plot(kind="scatter", x="rooms_per_household", y="median_house_value",
alpha=0.2)
plt.axis([0, 5, 0, 520000])
plt.show()
housing.describe()
Prepare the data for Machine Learning algorithms¶
housing = strat_train_set.drop("median_house_value", axis=1) # drop labels for training set
housing_labels = strat_train_set["median_house_value"].copy()
sample_incomplete_rows = housing[housing.isnull().any(axis=1)].head()
sample_incomplete_rows
sample_incomplete_rows.dropna(subset=["total_bedrooms"]) # option 1
sample_incomplete_rows.drop("total_bedrooms", axis=1) # option 2
median = housing["total_bedrooms"].median()
sample_incomplete_rows["total_bedrooms"].fillna(median, inplace=True) # option 3
sample_incomplete_rows
from sklearn.preprocessing import Imputer
imputer = Imputer(strategy="median")
Remove the text attribute because median can only be calculated on numerical attributes:
housing_num = housing.drop('ocean_proximity', axis=1)
# alternatively: housing_num = housing.select_dtypes(include=[np.number])
imputer.fit(housing_num)
imputer.statistics_
Check that this is the same as manually computing the median of each attribute:
housing_num.median().values
Transform the training set:
X = imputer.transform(housing_num)
housing_tr = pd.DataFrame(X, columns=housing_num.columns,
index = list(housing.index.values))
housing_tr.loc[sample_incomplete_rows.index.values]
imputer.strategy
housing_tr = pd.DataFrame(X, columns=housing_num.columns)
housing_tr.head()
Now let's preprocess the categorical input feature, ocean_proximity:
housing_cat = housing[['ocean_proximity']]
housing_cat.head(10)
Warning: earlier versions of the book used the LabelEncoder class or Pandas' Series.factorize() method to encode string categorical attributes as integers. However, the OrdinalEncoder class that is planned to be introduced in Scikit-Learn 0.20 (see PR #10521) is preferable since it is designed for input features (X instead of labels y) and it plays well with pipelines (introduced later in this notebook). For now, we will import it from future_encoders.py, but once it is available you can import it directly from sklearn.preprocessing.
from future_encoders import OrdinalEncoder
ordinal_encoder = OrdinalEncoder()
housing_cat_encoded = ordinal_encoder.fit_transform(housing_cat)
housing_cat_encoded[:10]
ordinal_encoder.categories_
Warning: earlier versions of the book used the LabelBinarizer or CategoricalEncoder classes to convert each categorical value to a one-hot vector. It is now preferable to use the OneHotEncoder class. Right now it can only handle integer categorical inputs, but in Scikit-Learn 0.20 it will also handle string categorical inputs (see PR #10521). So for now we import it from future_encoders.py, but when Scikit-Learn 0.20 is released, you can import it from sklearn.preprocessing instead:
from future_encoders import OneHotEncoder
cat_encoder = OneHotEncoder()
housing_cat_1hot = cat_encoder.fit_transform(housing_cat)
housing_cat_1hot
By default, the OneHotEncoder class returns a sparse array, but we can convert it to a dense array if needed by calling the toarray() method:
housing_cat_1hot.toarray()
Alternatively, you can set sparse=False when creating the OneHotEncoder:
cat_encoder = OneHotEncoder(sparse=False)
housing_cat_1hot = cat_encoder.fit_transform(housing_cat)
housing_cat_1hot
cat_encoder.categories_
Let's create a custom transformer to add extra attributes:
from sklearn.base import BaseEstimator, TransformerMixin
# column index
rooms_ix, bedrooms_ix, population_ix, household_ix = 3, 4, 5, 6
class CombinedAttributesAdder(BaseEstimator, TransformerMixin):
def __init__(self, add_bedrooms_per_room = True): # no *args or **kargs
self.add_bedrooms_per_room = add_bedrooms_per_room
def fit(self, X, y=None):
return self # nothing else to do
def transform(self, X, y=None):
rooms_per_household = X[:, rooms_ix] / X[:, household_ix]
population_per_household = X[:, population_ix] / X[:, household_ix]
if self.add_bedrooms_per_room:
bedrooms_per_room = X[:, bedrooms_ix] / X[:, rooms_ix]
return np.c_[X, rooms_per_household, population_per_household,
bedrooms_per_room]
else:
return np.c_[X, rooms_per_household, population_per_household]
attr_adder = CombinedAttributesAdder(add_bedrooms_per_room=False)
housing_extra_attribs = attr_adder.transform(housing.values)
housing_extra_attribs = pd.DataFrame(
housing_extra_attribs,
columns=list(housing.columns)+["rooms_per_household", "population_per_household"])
housing_extra_attribs.head()
Now let's build a pipeline for preprocessing the numerical attributes:
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
num_pipeline = Pipeline([
('imputer', Imputer(strategy="median")),
('attribs_adder', CombinedAttributesAdder()),
('std_scaler', StandardScaler()),
])
housing_num_tr = num_pipeline.fit_transform(housing_num)
housing_num_tr
Warning: earlier versions of the book applied different transformations to different columns using a solution based on a DataFrameSelector transformer and a FeatureUnion (see below). It is now preferable to use the ColumnTransformer class that will be introduced in Scikit-Learn 0.20. For now we import it from future_encoders.py, but when Scikit-Learn 0.20 is released, you can import it from sklearn.compose instead:
from future_encoders import ColumnTransformer
num_attribs = list(housing_num)
cat_attribs = ["ocean_proximity"]
full_pipeline = ColumnTransformer([
("num", num_pipeline, num_attribs),
("cat", OneHotEncoder(), cat_attribs),
])
housing_prepared = full_pipeline.fit_transform(housing)
housing_prepared
housing_prepared.shape
For reference, here is the old solution based on a DataFrameSelector transformer (to just select a subset of the Pandas DataFrame columns), and a FeatureUnion:
from sklearn.base import BaseEstimator, TransformerMixin
# Create a class to select numerical or categorical columns
class OldDataFrameSelector(BaseEstimator, TransformerMixin):
def __init__(self, attribute_names):
self.attribute_names = attribute_names
def fit(self, X, y=None):
return self
def transform(self, X):
return X[self.attribute_names].values
Now let's join all these components into a big pipeline that will preprocess both the numerical and the categorical features:
num_attribs = list(housing_num)
cat_attribs = ["ocean_proximity"]
old_num_pipeline = Pipeline([
('selector', OldDataFrameSelector(num_attribs)),
('imputer', Imputer(strategy="median")),
('attribs_adder', CombinedAttributesAdder()),
('std_scaler', StandardScaler()),
])
old_cat_pipeline = Pipeline([
('selector', OldDataFrameSelector(cat_attribs)),
('cat_encoder', OneHotEncoder(sparse=False)),
])
from sklearn.pipeline import FeatureUnion
old_full_pipeline = FeatureUnion(transformer_list=[
("num_pipeline", old_num_pipeline),
("cat_pipeline", old_cat_pipeline),
])
old_housing_prepared = old_full_pipeline.fit_transform(housing)
old_housing_prepared
The result is the same as with the ColumnTransformer:
np.allclose(housing_prepared, old_housing_prepared)
Select and train a model¶
from sklearn.linear_model import LinearRegression
lin_reg = LinearRegression()
lin_reg.fit(housing_prepared, housing_labels)
# let's try the full preprocessing pipeline on a few training instances
some_data = housing.iloc[:5]
some_labels = housing_labels.iloc[:5]
some_data_prepared = full_pipeline.transform(some_data)
print("Predictions:", lin_reg.predict(some_data_prepared))
Compare against the actual values:
print("Labels:", list(some_labels))
some_data_prepared
from sklearn.metrics import mean_squared_error
housing_predictions = lin_reg.predict(housing_prepared)
lin_mse = mean_squared_error(housing_labels, housing_predictions)
lin_rmse = np.sqrt(lin_mse)
lin_rmse
from sklearn.metrics import mean_absolute_error
lin_mae = mean_absolute_error(housing_labels, housing_predictions)
lin_mae
from sklearn.tree import DecisionTreeRegressor
tree_reg = DecisionTreeRegressor(random_state=42)
tree_reg.fit(housing_prepared, housing_labels)
housing_predictions = tree_reg.predict(housing_prepared)
tree_mse = mean_squared_error(housing_labels, housing_predictions)
tree_rmse = np.sqrt(tree_mse)
tree_rmse
Fine-tune your model¶
from sklearn.model_selection import cross_val_score
scores = cross_val_score(tree_reg, housing_prepared, housing_labels,
scoring="neg_mean_squared_error", cv=10)
tree_rmse_scores = np.sqrt(-scores)
def display_scores(scores):
print("Scores:", scores)
print("Mean:", scores.mean())
print("Standard deviation:", scores.std())
display_scores(tree_rmse_scores)
lin_scores = cross_val_score(lin_reg, housing_prepared, housing_labels,
scoring="neg_mean_squared_error", cv=10)
lin_rmse_scores = np.sqrt(-lin_scores)
display_scores(lin_rmse_scores)
from sklearn.ensemble import RandomForestRegressor
forest_reg = RandomForestRegressor(random_state=42)
forest_reg.fit(housing_prepared, housing_labels)
housing_predictions = forest_reg.predict(housing_prepared)
forest_mse = mean_squared_error(housing_labels, housing_predictions)
forest_rmse = np.sqrt(forest_mse)
forest_rmse
from sklearn.model_selection import cross_val_score
forest_scores = cross_val_score(forest_reg, housing_prepared, housing_labels,
scoring="neg_mean_squared_error", cv=10)
forest_rmse_scores = np.sqrt(-forest_scores)
display_scores(forest_rmse_scores)
scores = cross_val_score(lin_reg, housing_prepared, housing_labels, scoring="neg_mean_squared_error", cv=10)
pd.Series(np.sqrt(-scores)).describe()
from sklearn.svm import SVR
svm_reg = SVR(kernel="linear")
svm_reg.fit(housing_prepared, housing_labels)
housing_predictions = svm_reg.predict(housing_prepared)
svm_mse = mean_squared_error(housing_labels, housing_predictions)
svm_rmse = np.sqrt(svm_mse)
svm_rmse
from sklearn.model_selection import GridSearchCV
param_grid = [
# try 12 (3×4) combinations of hyperparameters
{'n_estimators': [3, 10, 30], 'max_features': [2, 4, 6, 8]},
# then try 6 (2×3) combinations with bootstrap set as False
{'bootstrap': [False], 'n_estimators': [3, 10], 'max_features': [2, 3, 4]},
]
forest_reg = RandomForestRegressor(random_state=42)
# train across 5 folds, that's a total of (12+6)*5=90 rounds of training
grid_search = GridSearchCV(forest_reg, param_grid, cv=5,
scoring='neg_mean_squared_error', return_train_score=True)
grid_search.fit(housing_prepared, housing_labels)
The best hyperparameter combination found:
grid_search.best_params_
grid_search.best_estimator_
Let's look at the score of each hyperparameter combination tested during the grid search:
cvres = grid_search.cv_results_
for mean_score, params in zip(cvres["mean_test_score"], cvres["params"]):
print(np.sqrt(-mean_score), params)
pd.DataFrame(grid_search.cv_results_)
from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import randint
param_distribs = {
'n_estimators': randint(low=1, high=200),
'max_features': randint(low=1, high=8),
}
forest_reg = RandomForestRegressor(random_state=42)
rnd_search = RandomizedSearchCV(forest_reg, param_distributions=param_distribs,
n_iter=10, cv=5, scoring='neg_mean_squared_error', random_state=42)
rnd_search.fit(housing_prepared, housing_labels)
cvres = rnd_search.cv_results_
for mean_score, params in zip(cvres["mean_test_score"], cvres["params"]):
print(np.sqrt(-mean_score), params)
feature_importances = grid_search.best_estimator_.feature_importances_
feature_importances
extra_attribs = ["rooms_per_hhold", "pop_per_hhold", "bedrooms_per_room"]
#cat_encoder = cat_pipeline.named_steps["cat_encoder"] # old solution
cat_encoder = full_pipeline.named_transformers_["cat"]
cat_one_hot_attribs = list(cat_encoder.categories_[0])
attributes = num_attribs + extra_attribs + cat_one_hot_attribs
sorted(zip(feature_importances, attributes), reverse=True)
final_model = grid_search.best_estimator_
X_test = strat_test_set.drop("median_house_value", axis=1)
y_test = strat_test_set["median_house_value"].copy()
X_test_prepared = full_pipeline.transform(X_test)
final_predictions = final_model.predict(X_test_prepared)
final_mse = mean_squared_error(y_test, final_predictions)
final_rmse = np.sqrt(final_mse)
final_rmse
We can compute a 95% confidence interval for the test RMSE:
from scipy import stats
confidence = 0.95
squared_errors = (final_predictions - y_test) ** 2
mean = squared_errors.mean()
m = len(squared_errors)
np.sqrt(stats.t.interval(confidence, m - 1,
loc=np.mean(squared_errors),
scale=stats.sem(squared_errors)))
We could compute the interval manually like this:
tscore = stats.t.ppf((1 + confidence) / 2, df=m - 1)
tmargin = tscore * squared_errors.std(ddof=1) / np.sqrt(m)
np.sqrt(mean - tmargin), np.sqrt(mean + tmargin)
Alternatively, we could use a z-scores rather than t-scores:
zscore = stats.norm.ppf((1 + confidence) / 2)
zmargin = zscore * squared_errors.std(ddof=1) / np.sqrt(m)
np.sqrt(mean - zmargin), np.sqrt(mean + zmargin)
Extra material¶
A full pipeline with both preparation and prediction¶
full_pipeline_with_predictor = Pipeline([
("preparation", full_pipeline),
("linear", LinearRegression())
])
full_pipeline_with_predictor.fit(housing, housing_labels)
full_pipeline_with_predictor.predict(some_data)
Model persistence using joblib¶
my_model = full_pipeline_with_predictor
from sklearn.externals import joblib
joblib.dump(my_model, "my_model.pkl") # DIFF
#...
my_model_loaded = joblib.load("my_model.pkl") # DIFF
Example SciPy distributions for RandomizedSearchCV¶
from scipy.stats import geom, expon
geom_distrib=geom(0.5).rvs(10000, random_state=42)
expon_distrib=expon(scale=1).rvs(10000, random_state=42)
plt.hist(geom_distrib, bins=50)
plt.show()
plt.hist(expon_distrib, bins=50)
plt.show()
Exercise solutions¶
1.¶
Question: Try a Support Vector Machine regressor (sklearn.svm.SVR), with various hyperparameters such as kernel="linear" (with various values for the C hyperparameter) or kernel="rbf" (with various values for the C and gamma hyperparameters). Don't worry about what these hyperparameters mean for now. How does the best SVR predictor perform?
from sklearn.model_selection import GridSearchCV
param_grid = [
{'kernel': ['linear'], 'C': [10., 30., 100., 300., 1000., 3000., 10000., 30000.0]},
{'kernel': ['rbf'], 'C': [1.0, 3.0, 10., 30., 100., 300., 1000.0],
'gamma': [0.01, 0.03, 0.1, 0.3, 1.0, 3.0]},
]
svm_reg = SVR()
grid_search = GridSearchCV(svm_reg, param_grid, cv=5, scoring='neg_mean_squared_error', verbose=2, n_jobs=4)
grid_search.fit(housing_prepared, housing_labels)
The best model achieves the following score (evaluated using 5-fold cross validation):
negative_mse = grid_search.best_score_
rmse = np.sqrt(-negative_mse)
rmse
That's much worse than the RandomForestRegressor. Let's check the best hyperparameters found:
grid_search.best_params_
The linear kernel seems better than the RBF kernel. Notice that the value of C is the maximum tested value. When this happens you definitely want to launch the grid search again with higher values for C (removing the smallest values), because it is likely that higher values of C will be better.
2.¶
Question: Try replacing GridSearchCV with RandomizedSearchCV.
from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import expon, reciprocal
# see https://docs.scipy.org/doc/scipy/reference/stats.html
# for `expon()` and `reciprocal()` documentation and more probability distribution functions.
# Note: gamma is ignored when kernel is "linear"
param_distribs = {
'kernel': ['linear', 'rbf'],
'C': reciprocal(20, 200000),
'gamma': expon(scale=1.0),
}
svm_reg = SVR()
rnd_search = RandomizedSearchCV(svm_reg, param_distributions=param_distribs,
n_iter=50, cv=5, scoring='neg_mean_squared_error',
verbose=2, n_jobs=4, random_state=42)
rnd_search.fit(housing_prepared, housing_labels)
The best model achieves the following score (evaluated using 5-fold cross validation):
negative_mse = rnd_search.best_score_
rmse = np.sqrt(-negative_mse)
rmse
Now this is much closer to the performance of the RandomForestRegressor (but not quite there yet). Let's check the best hyperparameters found:
rnd_search.best_params_
This time the search found a good set of hyperparameters for the RBF kernel. Randomized search tends to find better hyperparameters than grid search in the same amount of time.
Let's look at the exponential distribution we used, with scale=1.0. Note that some samples are much larger or smaller than 1.0, but when you look at the log of the distribution, you can see that most values are actually concentrated roughly in the range of exp(-2) to exp(+2), which is about 0.1 to 7.4.
expon_distrib = expon(scale=1.)
samples = expon_distrib.rvs(10000, random_state=42)
plt.figure(figsize=(10, 4))
plt.subplot(121)
plt.title("Exponential distribution (scale=1.0)")
plt.hist(samples, bins=50)
plt.subplot(122)
plt.title("Log of this distribution")
plt.hist(np.log(samples), bins=50)
plt.show()
The distribution we used for C looks quite different: the scale of the samples is picked from a uniform distribution within a given range, which is why the right graph, which represents the log of the samples, looks roughly constant. This distribution is useful when you don't have a clue of what the target scale is:
reciprocal_distrib = reciprocal(20, 200000)
samples = reciprocal_distrib.rvs(10000, random_state=42)
plt.figure(figsize=(10, 4))
plt.subplot(121)
plt.title("Reciprocal distribution (scale=1.0)")
plt.hist(samples, bins=50)
plt.subplot(122)
plt.title("Log of this distribution")
plt.hist(np.log(samples), bins=50)
plt.show()
The reciprocal distribution is useful when you have no idea what the scale of the hyperparameter should be (indeed, as you can see on the figure on the right, all scales are equally likely, within the given range), whereas the exponential distribution is best when you know (more or less) what the scale of the hyperparameter should be.
3.¶
Question: Try adding a transformer in the preparation pipeline to select only the most important attributes.
from sklearn.base import BaseEstimator, TransformerMixin
def indices_of_top_k(arr, k):
return np.sort(np.argpartition(np.array(arr), -k)[-k:])
class TopFeatureSelector(BaseEstimator, TransformerMixin):
def __init__(self, feature_importances, k):
self.feature_importances = feature_importances
self.k = k
def fit(self, X, y=None):
self.feature_indices_ = indices_of_top_k(self.feature_importances, self.k)
return self
def transform(self, X):
return X[:, self.feature_indices_]
Note: this feature selector assumes that you have already computed the feature importances somehow (for example using a RandomForestRegressor). You may be tempted to compute them directly in the TopFeatureSelector's fit() method, however this would likely slow down grid/randomized search since the feature importances would have to be computed for every hyperparameter combination (unless you implement some sort of cache).
Let's define the number of top features we want to keep:
k = 5
Now let's look for the indices of the top k features:
top_k_feature_indices = indices_of_top_k(feature_importances, k)
top_k_feature_indices
np.array(attributes)[top_k_feature_indices]
Let's double check that these are indeed the top k features:
sorted(zip(feature_importances, attributes), reverse=True)[:k]
Looking good... Now let's create a new pipeline that runs the previously defined preparation pipeline, and adds top k feature selection:
preparation_and_feature_selection_pipeline = Pipeline([
('preparation', full_pipeline),
('feature_selection', TopFeatureSelector(feature_importances, k))
])
housing_prepared_top_k_features = preparation_and_feature_selection_pipeline.fit_transform(housing)
Let's look at the features of the first 3 instances:
housing_prepared_top_k_features[0:3]
Now let's double check that these are indeed the top k features:
housing_prepared[0:3, top_k_feature_indices]
Works great! :)
4.¶
Question: Try creating a single pipeline that does the full data preparation plus the final prediction.
prepare_select_and_predict_pipeline = Pipeline([
('preparation', full_pipeline),
('feature_selection', TopFeatureSelector(feature_importances, k)),
('svm_reg', SVR(**rnd_search.best_params_))
])
prepare_select_and_predict_pipeline.fit(housing, housing_labels)
Let's try the full pipeline on a few instances:
some_data = housing.iloc[:4]
some_labels = housing_labels.iloc[:4]
print("Predictions:\t", prepare_select_and_predict_pipeline.predict(some_data))
print("Labels:\t\t", list(some_labels))
Well, the full pipeline seems to work fine. Of course, the predictions are not fantastic: they would be better if we used the best RandomForestRegressor that we found earlier, rather than the best SVR.
5.¶
Question: Automatically explore some preparation options using GridSearchCV.
param_grid = [
{'preparation__num__imputer__strategy': ['mean', 'median', 'most_frequent'],
'feature_selection__k': list(range(1, len(feature_importances) + 1))}
]
grid_search_prep = GridSearchCV(prepare_select_and_predict_pipeline, param_grid, cv=5,
scoring='neg_mean_squared_error', verbose=2, n_jobs=4)
grid_search_prep.fit(housing, housing_labels)
grid_search_prep.best_params_
The best imputer strategy is most_frequent and apparently almost all features are useful (15 out of 16). The last one (ISLAND) seems to just add some noise.
Congratulations! You already know quite a lot about Machine Learning. :)