def test_on_housing_dataset(self):
"""
Test on housing data set
Logging results
:return: None
"""
log = logging.getLogger("DecisionTreeTest.test_on_housing_dataset")
data = np.loadtxt(DecisionTreeTest.HousingDataPath)
x, y = data[::, :-1:], data[::, -1]
kf = KFold(x.shape[0], n_folds=5)
for train, test in kf:
train_x, train_y = x[train], y[train]
test_x, test_y = x[test], y[test]
tree = DecisionTreeRegressor(max_depth=50, min_list_size=2, min_list_variance=1e-5)
tree.fit(train_x, train_y)
sktree = sklearn_trees.DecisionTreeRegressor()
sktree.fit(train_x, train_y)
prediction = tree.predict(test_x)
skprediction = sktree.predict(test_x)
log.debug("Target: %s" % test_y)
log.debug("Prediction: %s" % prediction)
log.debug("Mean squared error my tree: %f" % mean_squared_error(test_y, prediction))
log.debug("Mean squared error sklearn tree: %f" % mean_squared_error(test_y, skprediction))
def test_tree(filename):
df = pd.read_csv(filename)
y = df.pop('Humidity').values
X = df.values
print X
tree = DecisionTreeRegressor()
tree.fit(X, y, df.columns)
tree.prune(X, y)
print tree
y_predict = tree.predict(X)
print '%35s %10s %10s' % ("FEATURES", "ACTUAL", "PREDICTED")
print '%35s %10s %10s' % ("----------", "----------", "----------")
for features, true, predicted in izip(X, y, y_predict):
print '%35s %10d %10d' % (str(features), true, predicted)
def test_on_test_dataset(self):
"""
Test on test data set
Logging results
:return: None
"""
log = logging.getLogger("DecisionTreeTest.test_on_test_dataset")
tree = DecisionTreeRegressor()
data = np.loadtxt(DecisionTreeTest.TestDataPath)
tree.fit(data[::, :-1:], data[::, -1])
prediction = tree.predict(data[::, :-1:])
y = data[::, -1]
log.debug("Prediction: {0}".format(prediction))
log.debug("Target value: {0}".format(y))
self.assertTrue(np.array_equal(prediction, y))
def fit(self, train, target=None, test=None):
"""
Fit the random forest to the training set train. If a test set is provided
then the return value wil be the predictions of the RandomForest on the
test set. If no test set is provide nothing is returned.
Note: Below we set the number of features to use in the splitting to be
the square root of the number of total features in the dataset.
:Parameters:
**train** (list or `Pandas DataFrame <http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.html>`_) : The training set.
**target** (str or None) : The name of the target variable
**test** (list or `Pandas DataFrame <http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.html>`_) : The test set.
:Returns:
(list or None): If a test set is provided then the return value wil be
the predictions of the RandomForest on the test set. If no test set
is provide nothing is returned.
"""
# set the number of features for the trees to use.
if isinstance(train, list) is False:
if target is None:
raise ValueError(
'If passing dataframe need to specify target.')
else:
train = self._convert_dataframe_to_list(train, target)
n_features = int(sqrt(len(train[0]) - 1))
for i in range(self.n_trees):
sample = self._subsample(train)
tree = DecisionTreeRegressor(self.max_depth, self.min_size,
self.cost_function)
tree.fit(sample, n_features)
self.trees.append(tree)
# if the test set is not empty then return the predictions
if test is not None:
predictions = [self.predict(row) for row in test]
return (predictions)
def fit(self, train, target=None, test=None):
# set the number of features for the trees to use.
if isinstance(train, list) is False:
if target is None:
raise ValueError('If passing dataframe need to specify target.')
else:
train = self._convert_dataframe_to_list(train, target)
n_features = int(sqrt(len(train[0])-1))
for i in range(self.n_trees):
sample = self._subsample(train)
tree = DecisionTreeRegressor(self.max_depth, self.min_size, self.cost_function)
tree.fit(sample, n_features)
self.trees.append(tree)
# if the test set is not empty then return the predictions
if test is not None:
predictions = [self.predict(row) for row in test]
return(predictions)
