class DecisionTreeClassifierImpl():
def __init__(self, criterion='gini', splitter='best', max_depth=None, min_samples_split=2, min_samples_leaf=1, min_weight_fraction_leaf=0.0, max_features=None, random_state=None, max_leaf_nodes=None, min_impurity_decrease=0.0, min_impurity_split=None, class_weight='balanced', presort=False):
self._hyperparams = {
'criterion': criterion,
'splitter': splitter,
'max_depth': max_depth,
'min_samples_split': min_samples_split,
'min_samples_leaf': min_samples_leaf,
'min_weight_fraction_leaf': min_weight_fraction_leaf,
'max_features': max_features,
'random_state': random_state,
'max_leaf_nodes': max_leaf_nodes,
'min_impurity_decrease': min_impurity_decrease,
'min_impurity_split': min_impurity_split,
'class_weight': class_weight,
'presort': presort}
def fit(self, X, y=None):
self._sklearn_model = SKLModel(**self._hyperparams)
if (y is not None):
self._sklearn_model.fit(X, y)
else:
self._sklearn_model.fit(X)
return self
def predict(self, X):
return self._sklearn_model.predict(X)
def predict_proba(self, X):
return self._sklearn_model.predict_proba(X)
def fit(self, X, y):
"""
Fit the NearestCentroid model according to the given training data.
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training vector, where n_samples in the number of samples and
n_features is the number of features.
Note that centroid shrinking cannot be used with sparse matrices.
y : array, shape = [n_samples]
Target values (integers)
"""
self.y = y
if self.fit_base:
self.base_classifier.fit(X, y)
distances = self.base_classifier.predict_proba(X)
topNIndices, topNDistances = self._get_top_labels(distances)
training_data = self._extract_features(topNIndices, topNDistances, y,
distances)
# create a decision tree for each label
self.meta_classifiers = {}
for label, training_samples_of_label in training_data.items():
training_samples_of_label = np.matrix(training_samples_of_label)
decision_tree = DecisionTreeClassifier(criterion="gini")
decision_tree.fit(training_samples_of_label[:, 0:-1],
training_samples_of_label[:, -1:])
self.meta_classifiers[label] = decision_tree
def decision_tree_depths():
max_depths = [2, 4, 6, 8, 10, 12, 16, 18, 20, 25, 30, 40]
columns = [
'Max Depths', 'Training Score', 'Test Score', 'Train Time', 'Test Time'
]
df = pd.DataFrame(columns=columns)
for depth in max_depths:
start_train = time.time()
dt = DecisionTreeClassifier(max_depth=depth)
print(dt)
dt.fit(X_train, y_train)
end_train = time.time() - start_train
train_score = dt.score(X_train, y_train)
start_test = time.time()
test_score = dt.score(X_test, y_test)
end_test = time.time() - start_test
values = [depth, train_score, test_score, end_train, end_test]
df.loc[len(df)] = values
print(' '.join(str(col) for col in columns))
print(' '.join(str(val) for val in values))
df.to_excel('adult_dt.xls')
def decision_tree_training_sets():
training_set_sizes = [.1,.25,.5,.75,.9]
columns = ['Training Set Size', 'Training Score', 'Test Score', 'Train Time', 'Test Time']
df = pd.DataFrame(columns=columns)
for training_set_size in training_set_sizes:
X_train, X_test, y_train, y_test = train_test_split(
encoded_data[list(set(encoded_data.columns) - set(['Target']))],
encoded_data['Target'], train_size=training_set_size)
scaler = preprocessing.StandardScaler()
X_train = pd.DataFrame(scaler.fit_transform(X_train.astype('float32')), columns=X_train.columns)
X_test = scaler.transform(X_test.astype('float32'))
start_train = time.time()
dt = DecisionTreeClassifier(max_depth=8)
print(dt)
dt.fit(X_train, y_train)
end_train = time.time() - start_train
train_score = dt.score(X_train, y_train)
start_test = time.time()
test_score = dt.score(X_test, y_test)
end_test = time.time() - start_test
values = [training_set_size, train_score, test_score, end_train, end_test]
df.loc[len(df)] = values
print(' '.join(str(col) for col in columns))
print(' '.join(str(val) for val in values))
df.to_excel('diabetes_dt_training_sets.xls')
def wrapper_for_decision_tree_in_sklearn(X, y, current_state_to_predict):
clf = DecisionTreeClassifier()
clf.fit(X, y)
current_state_to_predict = np.array(current_state_to_predict).reshape(
1, -1)
predicted_state = clf.predict(current_state_to_predict)
return predicted_state
def fit(self, X, y):
"""
Fit the NearestCentroid model according to the given training data.
Parameters
----------
X : {array-like, sparse matrix}, shape = [n_samples, n_features]
Training vector, where n_samples in the number of samples and
n_features is the number of features.
Note that centroid shrinking cannot be used with sparse matrices.
y : array, shape = [n_samples]
Target values (integers)
"""
self.y = y
if self.fit_base:
self.base_classifier.fit(X, y)
distances = self.base_classifier.predict_proba(X)
topNIndices, topNDistances = self._get_top_labels(distances)
training_data = self._extract_features(topNIndices, topNDistances, y, distances)
# create a decision tree for each label
self.meta_classifiers = {}
for label, training_samples_of_label in training_data.items():
training_samples_of_label = np.matrix(training_samples_of_label)
decision_tree = DecisionTreeClassifier(criterion="gini")
decision_tree.fit(training_samples_of_label[:, 0:-1], training_samples_of_label[:, -1:])
self.meta_classifiers[label] = decision_tree
def dtree(X, y, model_path):
model = DecisionTreeClassifier()
model.fit(X, y)
expected = y
predicted = model.predict(X)
print(metrics.classification_report(expected, predicted))
print(metrics.confusion_matrix(expected, predicted))
joblib.dump(model, model_path)
def wrapper_for_decision_tree_accuracy(X, y, relative_test_size):
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=relative_test_size, random_state=42)
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
score = accuracy_score(pred, y_test)
return score
def wrapper_for_decision_tree_accuracy(X, y, relative_test_size):
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=relative_test_size, random_state=42)
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
score = accuracy_score(pred,y_test)
return score
def create_decision_tree(self):
''' based on experiments our best model was the decision tree model with the following params: '''
tree = DecisionTreeClassifier(max_depth=65,
min_samples_split=0.03,
min_samples_leaf=3,
max_features=8)
tree.fit(self.X_train, self.Y_train)
predicted_y = tree.predict(self.X_test)
print(predicted_y)
self.print_stats(predicted_y, "")
self.test_df['learning_label'] = predicted_y
self.test_df.to_csv('output/feature_extraction.csv',
encoding="latin-1") # save the training dataset
def MarginBoostClf(features, labels, max_depth, n_steps, margin):
sample_size = features.shape[0]
weights = np.ones(sample_size) / sample_size
clf_list = []
for t in range(n_steps):
clf = DecisionTreeClassifier(max_depth=max_depth)
clf = clf.fit(features, labels, sample_weight=weights)
y_predict = clf.predict(features)
incorrect = y_predict != labels
# Error fraction
estimator_error = np.mean(
np.average(incorrect, weights=weights, axis=0))
if (estimator_error >= 0.5):
break
step_size = 0.5 * (np.log((1 - estimator_error) / estimator_error) +
np.log(1 - margin) - np.log(1 + margin))
norm_factor = 2 * pow(estimator_error * (1 - estimator_error), 0.5)
for i in range(sample_size):
if (labels[i] == y_predict[i]):
weights[i] *= np.exp(-step_size) / norm_factor
else:
weights[i] *= np.exp(step_size) / norm_factor
clf_list.append([clf, step_size])
return clf_list
def BoostByMaj(features, labels, max_depth, gamma):
sample_size = features.shape[0]
weights = np.ones(sample_size) / sample_size
counts = np.zeros(sample_size)
k_pre = get_k_from_gamma(gamma, sample_size)
k = k_pre
#k = min(600, k_pre)
print('k ', k)
clf_list = []
for i in range(k):
estimator_error = 0.6
countdown = 10
while ((estimator_error >= 0.5) and (countdown >= 0)):
clf = DecisionTreeClassifier(max_depth=max_depth)
clf = clf.fit(features, labels, sample_weight=weights)
y_predict = clf.predict(features)
correct_ones = y_predict == labels
incorrect_ones = y_predict != labels
estimator_error = np.mean(
np.average(incorrect_ones, weights=weights, axis=0))
unweighted_estimator_error = np.mean(
np.average(incorrect_ones, axis=0))
countdown -= 1
counts += correct_ones
coeff_1 = int(np.floor(k / 2)) - counts
coeff_2 = int(np.ceil(k / 2)) - i - 1 + counts
weights = comb(k - i - 1, coeff_1) * pow(0.5 + gamma, coeff_1) * pow(
0.5 - gamma, coeff_2)
print('i', i, 'error', estimator_error, 'unweighted_error',
unweighted_estimator_error, 'wnorm',
np.linalg.norm(weights, ord=1))
weights = weights / np.linalg.norm(weights, ord=1)
clf_list.append([clf, 1])
return clf_list, weights
def sklearn_titanic():
from sklearn.tree.tree import DecisionTreeClassifier
from sklearn.preprocessing.label import LabelEncoder
total_df = pd.read_csv("titanic_clean.csv")
total_df.drop(['cabin', 'boat', 'body', 'index'], axis=1, inplace=True)
total_df.dropna(inplace=True)
for col in total_df.columns.tolist():
if str(total_df[col].dtype) == 'object':
total_df[col] = LabelEncoder().fit_transform(total_df[col])
total_num = total_df.shape[0]
train_df = total_df.iloc[:int(total_num * 0.8)]
test_df = total_df.iloc[int(total_num * 0.8):]
clf = DecisionTreeClassifier()
clf.fit(train_df.drop(['survived'], axis=1), train_df['survived'])
print(clf.score(test_df.drop(['survived'], axis=1), test_df['survived']))
def DeepBBM2(features, labels, max_depth, gamma, max_depth_range):
num_features = features.shape[1]
sample_size = features.shape[0]
weights = np.ones(sample_size) / sample_size
D_weights = np.ones(sample_size) / sample_size
counts = np.zeros(sample_size)
k_pre = get_k_from_gamma(gamma, sample_size)
k = k_pre
#k = min(600, k_pre)
normalizer = np.exp(1) * sample_size
print('k ', k)
clf_list = []
rademacher_list = []
for depth in max_depth_range:
rademacher_list.append(
calc_rademacher(depth, sample_size, num_features, normalizer))
for t in range(k):
best_loss = 10000
best_error = 1
best_depth = -1
best_clf = DecisionTreeClassifier(max_depth=0)
for depth in max_depth_range:
new_clf_list, new_weights = DeepBoost(features,
labels,
1,
max_depth_range,
initial_weights=weights)
new_clf = DecisionTreeClassifier(max_depth=depth)
new_clf = new_clf.fit(features, labels, sample_weight=weights)
new_error = eval_clf(new_clf, features, labels, weights)
new_edge = new_error - 0.5
new_sign_edge = np.sign(new_edge)
new_loss = new_error + PARAM_lambda_2 * rademacher_list[depth - 1]
# print ('new_error', new_error, 'new_grad', new_grad)
print('depth', depth, 'new_error', new_error, 'new_grad', new_loss)
if (new_loss < best_loss):
best_clf = new_clf
best_loss = new_loss
best_error = new_error
best_depth = depth
y_predict = best_clf.predict(features)
correct_ones = y_predict == labels
counts += correct_ones
# if (best_error >= 0.5):
# break;
coeff_1 = int(np.floor(k / 2)) - counts
coeff_2 = int(np.ceil(k / 2)) - t - 1 + counts
weights = comb(k - t - 1, coeff_1) * pow(0.5 + gamma, coeff_1) * pow(
0.5 - gamma, coeff_2)
print('i', t, 'error', best_error, 'wnorm',
np.linalg.norm(weights, ord=1))
weights = weights / np.linalg.norm(weights, ord=1)
clf_list.append([best_clf, 1])
return clf_list, weights
def DeepBBM(features, labels, gamma, max_depth_range, PARAM_lambda_2):
num_features = features.shape[1]
sample_size = features.shape[0]
weights = np.ones(sample_size) / sample_size
counts = np.zeros(sample_size)
k_pre = get_k_from_gamma(gamma, sample_size)
k = k_pre
#k = min(600, k_pre)
normalizer = np.exp(1) * sample_size
# print ('k ', k)
clf_list = []
rademacher_list = []
for depth_index in range(len(max_depth_range)):
depth = max_depth_range[depth_index]
rademacher_list.append(
calc_rademacher(depth, sample_size, num_features, normalizer))
for t in range(k):
best_loss = 10000
best_error = 1
best_depth = -1
best_clf = DecisionTreeClassifier(max_depth=0)
for depth_index in range(len(max_depth_range)):
depth = max_depth_range[depth_index]
new_clf = DecisionTreeClassifier(max_depth=depth)
new_clf = new_clf.fit(features, labels, sample_weight=weights)
new_error = eval_clf(new_clf, features, labels, weights)
new_edge = new_error - 0.5
new_sign_edge = np.sign(new_edge)
new_loss = new_error + PARAM_lambda_2 * rademacher_list[depth_index]
if (new_loss < best_loss):
best_clf = new_clf
best_loss = new_loss
best_error = new_error
best_depth = depth
y_predict = best_clf.predict(features)
correct_ones = y_predict == labels
counts += correct_ones
coeff_1 = int(np.floor(k / 2)) - counts
coeff_2 = int(np.ceil(k / 2)) - t - 1 + counts
weights = comb(k - t - 1, coeff_1) * pow(0.5 + gamma, coeff_1) * pow(
0.5 - gamma, coeff_2)
clf_list.append([best_clf, 1, best_depth])
# print ('i', t, 'error', best_error, 'wnorm', np.linalg.norm(weights, ord=1))
if (np.max(coeff_1) < 0):
break
weights = weights / np.linalg.norm(weights, ord=1)
return clf_list, weights
def BrownBoost(features, labels, max_depth, total_time):
sample_size = features.shape[0]
clf_list = []
r = np.zeros(sample_size)
weights = np.array([])
s = total_time
#s works as the remaining time with the initial value total_time
T = total_time
alpha = 0
i = 0
b = np.zeros(sample_size)
while (s > 0 and i < 200):
weights = np.exp(-(r + s)**2 / total_time)
weights = weights / (np.sum(weights))
clf = DecisionTreeClassifier(max_depth=max_depth)
clf = clf.fit(features, labels, sample_weight=weights)
y_predict = clf.predict(features)
incorrect = y_predict != labels
# Error fraction
estimator_error = np.mean(
np.average(incorrect, weights=weights, axis=0))
print('estimator_error is', estimator_error)
if (estimator_error >= 0.5):
break
for j in range(sample_size):
if (labels[j] == y_predict[j]):
b[j] = 1
else:
b[j] = -1
a = r + s
(t, alpha) = SolveODE(a, b, s, sample_size, T)
r += alpha * b
s = s - t
print(s)
clf_list.append([clf, alpha])
i += 1
return clf_list
def ret_trained_DT_clf(X, Y):
clf = DecisionTreeClassifier(max_depth=3)
clf.fit(X, Y)
return clf
target_b = [0 if target[i] == "ELK" else 1 for i in range(len(target))]
target_c = [0 if target[i] == "CATTLE" else 1 for i in range(len(target))]
X_train_deer, X_test_deer, y_train_deer, y_test_deer = train_test_split(
train, target_a, random_state=0, test_size=0.3)
X_train_elk, X_test_elk, y_train_elk, y_test_elk = train_test_split(
train, target_b, random_state=0, test_size=0.3)
X_train_cattle, X_test_cattle, y_train_cattle, y_test_cattle = train_test_split(
train, target_c, random_state=0, test_size=0.3)
print("-----Question 1-----")
##Question 1##
###Decision Tree
print("-----DECISION TREE-----")
print("DEER confusion Matrix and accuracy score")
clf = DecisionTreeClassifier()
clf.fit(X_train_deer, y_train_deer)
y_pred = clf.predict(X_test_deer) ##predict my y's based on x's
print(confusion_matrix(y_test_deer, y_pred))
print("Testing Score")
print(accuracy_score(y_test_deer, y_pred)) #
y_pred = clf.predict(X_train_deer)
print("Training Score")
print(accuracy_score(y_train_deer, y_pred))
print("ELK confusion matrix and accuracy score")
clf = DecisionTreeClassifier()
clf.fit(X_train_elk, y_train_elk)
y_pred = clf.predict(X_test_elk)
print(confusion_matrix(y_test_elk, y_pred))
print("Testing Score")
print(accuracy_score(y_test_elk, y_pred))
Created on 2019年1月4日
决策树
'''
import numpy as np
from sklearn.model_selection._split import train_test_split
from sklearn.metrics.classification import classification_report
from sklearn.tree.tree import DecisionTreeClassifier
def iris_type(s):
it = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}
return it[str(s, encoding="utf8")]
path = 'demo1_Iris.txt' #
data = np.loadtxt(path, dtype=float, delimiter=',', converters={4: iris_type})
x, y = np.split(data, (4, ), axis=1)
x = x[:, :4]
x_train, x_test, y_train, y_test = train_test_split(x,
y,
random_state=1,
train_size=0.6)
clf = DecisionTreeClassifier(criterion='entropy', random_state=0)
clf.fit(x, y.ravel())
print('feature_importances_', clf.feature_importances_)
y_pred = clf.predict(x_test)
print(classification_report(y_test, y_pred))
class Model(object):
"""
The machine learning component of the tester.
This component stores four different models:
1) A model to decide between different types of events (drags and touches).
2) A model to decide on the starting position for drags.
3) A model to decide on the ending position for drags.
4) A model to decide on the position of the touch.
The input data are all the different known UI elements on the screen from
the training data and whether or not they are visible on the screen.
To acquire this, we first get the stored XML model and record the resource-id
and class. We concatenate them into an array and mark as (1) for visible and (0)
for not visible.
"""
def __init__(self):
self.symbols = {}
self.action_data = None
self.action_labels = None
self.action_classifier = None
self.drag_data = None
self.drag_end_labels = None
self.drag_end_classifier = None
self.drag_start_labels = None
self.drag_start_classifier = None
self.touch_data = None
self.touch_labels = None
self.touch_classifier = None
self.device_info = device.info
def parse_events(self, queue):
symbols = {"randomizer": 0}
events = []
all_data = []
all_results = []
drag_data = []
drag_start_results = []
drag_end_results = []
touch_data = []
touch_results = []
while not queue.empty():
event = queue.get()
events.append(event)
lst = event.state.start.as_list(symbols)
lst[0] = random()
all_data.append(lst)
if event.action.is_drag():
drag_data.append(lst)
all_results.append(DRAG)
start = event.changes.start()
end = event.changes.end()
drag_start_results.append(start.x * start.y)
drag_end_results.append(end.x * end.y)
if event.action.is_touch():
touch_data.append(lst)
all_results.append(TOUCH)
start = event.changes.start()
touch_results.append(start.x * start.y)
if event.action.is_back():
all_results.append(BACK)
data = np.zeros((len(all_data), len(symbols)))
for i, item in enumerate(all_data):
data[i, : len(item)] = item[:]
drags = np.zeros((len(drag_data), len(symbols)))
for i, item in enumerate(drag_data):
drags[i, : len(item)] = item[:]
touches = np.zeros((len(touch_data), len(symbols)))
for i, item in enumerate(touch_data):
touches[i, : len(item)] = item[:]
self.symbols = symbols
self.action_data = data
self.action_labels = np.array(all_results)
self.drag_data = drags
self.drag_start_labels = np.array(drag_start_results)
self.drag_end_labels = np.array(drag_end_results)
self.touch_data = touches
self.touch_labels = np.array(touch_results)
def train(self):
self.action_classifier = DecisionTreeClassifier()
self.action_classifier.fit(self.action_data, self.action_labels)
self.drag_start_classifier = DecisionTreeRegressor()
self.drag_start_classifier.fit(self.drag_data, self.drag_start_labels)
self.drag_end_classifier = DecisionTreeRegressor()
self.drag_end_classifier.fit(self.drag_data, self.drag_end_labels)
self.touch_classifier = DecisionTreeRegressor()
self.touch_classifier.fit(self.touch_data, self.touch_labels)
def predict(self, state):
input = state.as_list(self.symbols, False)
input[0] = random()
action = Action()
type = self.action_classifier.predict(input)
width = self.device_info["displayWidth"]
if type == DRAG:
start = self.drag_start_classifier.predict(input)[0]
end = self.drag_end_classifier.predict(input)[0]
start = Point(start % width, start / width)
end = Point(end % width, end / width)
action.init(ACTION_DRAG, start, end, 0.5)
elif type == TOUCH:
point = self.touch_classifier.predict(input)[0]
point = Point(point % width, point / width)
action.init(ACTION_TOUCH, point.x, point.y)
elif type == BACK:
action.init(ACTION_BACK)
return action
def save(self):
pass
def decision_tree_fit(X, y):
clf = DecisionTreeClassifier(min_samples_leaf=5, random_state=42)
return clf.fit(X, y)
build_iris(RidgeClassifierCV(), "RidgeIris", with_proba = False)
build_iris(BaggingClassifier(RidgeClassifier(random_state = 13), random_state = 13, n_estimators = 3, max_features = 0.5), "RidgeEnsembleIris")
build_iris(SGDClassifier(random_state = 13, max_iter = 100), "SGDIris", with_proba = False)
build_iris(SGDClassifier(random_state = 13, loss = "log", max_iter = 100), "SGDLogIris")
build_iris(SVC(), "SVCIris", with_proba = False)
build_iris(NuSVC(), "NuSVCIris", with_proba = False)
build_iris(VotingClassifier([("dt", DecisionTreeClassifier(random_state = 13)), ("nb", GaussianNB()), ("lr", LogisticRegression())]), "VotingEnsembleIris", with_proba = False)
build_iris(OptimalXGBClassifier(objective = "multi:softprob", ntree_limit = 7), "XGBIris", ntree_limit = 7)
if "Iris" in datasets:
mapper = DataFrameMapper([
(iris_X.columns.values, ContinuousDomain())
])
iris_Xt = mapper.fit_transform(iris_X)
dt_classifier = DecisionTreeClassifier(random_state = 13)
dt_classifier.fit(iris_Xt, iris_y)
lr_classifier = LogisticRegression(random_state = 13)
lr_classifier.fit(iris_Xt, iris_y)
pipeline = PMMLPipeline([
("mapper", mapper),
("estimator", SelectFirstEstimator([
("X[2] <= 3", dt_classifier),
(str(True), lr_classifier)
]))
])
pipeline.active_fields = iris_X.columns.values
pipeline.target_fields = ["Species"]
store_pkl(pipeline, "SelectFirstIris")
species = DataFrame(pipeline.predict(iris_X), columns = ["Species"])
store_csv(species, "SelectFirstIris")
### a classic way to overfit is to use a small number
### of data points and a large number of features;
### train on only 150 events to put ourselves in this regime
features_train = features_train[:150].toarray()
labels_train = labels_train[:150]
### your code goes here
from sklearn.tree.tree import DecisionTreeClassifier
vocab_list = vectorizer.get_feature_names()
dtc = DecisionTreeClassifier()
dtc.fit(features_train, labels_train)
pred = dtc.predict(features_test)
from sklearn.metrics import accuracy_score
accuracy = accuracy_score(labels_test, pred)
print(accuracy)
feature_importances = dtc.feature_importances_
for i in range(0, len(feature_importances)):
if feature_importances[i] > 0.2:
print("Importance = ", feature_importances[i], " number is ", i, " word is ", vocab_list[i])
def main():
print("Loading samples and labels")
samples, labels, _ = load_files("data")
print("Loaded {} samples".format(samples.shape[0]))
sequence_dim = 100
print("Converting to sequences of length {}".format(sequence_dim))
samples, labels = make_sequences(samples, labels, sequence_dim)
print("Number of samples from sequences: {}".format(samples.shape[0]))
lb = LabelBinarizer()
labels = lb.fit_transform(labels)
# flattened samples for Decision Tree
flatSamples = samples.reshape(samples.shape[0], -1) #tree!
(trainSamples, testSamples, trainLabels,
testLabels) = train_test_split(flatSamples,
labels,
test_size=0.25,
random_state=42)
print("=" * 20)
print("Building DecisionTree model")
model = DecisionTreeClassifier()
model.fit(trainSamples, trainLabels)
treeResults = model.predict(testSamples)
print(
confusion_matrix(testLabels.argmax(axis=1),
treeResults.argmax(axis=1)))
print(
classification_report(testLabels.argmax(axis=1),
treeResults.argmax(axis=1)))
treeAcc = accuracy_score(testLabels.argmax(axis=1),
treeResults.argmax(axis=1))
print("Accuracy Tree: {:.2f}".format(treeAcc))
print("Cohen's Kappa {:.2f}".format(
cohen_kappa_score(testLabels.argmax(axis=1),
treeResults.argmax(axis=1))))
print("=" * 20)
print("Building CNN model")
(trainSamples, testSamples, trainLabels,
testLabels) = train_test_split(samples,
labels,
test_size=0.25,
random_state=42)
inputShape = (samples.shape[1], samples.shape[2])
model = Sequential()
model.add(Conv1D(32, 10, padding="same", input_shape=inputShape))
model.add(Activation("relu"))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(Conv1D(64, 10, padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(Conv1D(128, 10, padding="same"))
model.add(Activation("relu"))
model.add(Dropout(0.2))
model.add(Flatten(input_shape=inputShape))
model.add(Dense(128, activation='sigmoid'))
model.add(Dense(64, activation='sigmoid'))
model.add(Dense(labels.shape[1], activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer="adam",
metrics=['accuracy'])
EPOCHS = 10
BATCH = 128
model.fit(trainSamples,
trainLabels,
batch_size=BATCH,
epochs=EPOCHS,
validation_data=(testSamples, testLabels))
cnnResults = model.predict(testSamples)
print(
confusion_matrix(testLabels.argmax(axis=1), cnnResults.argmax(axis=1)))
print(
classification_report(testLabels.argmax(axis=1),
cnnResults.argmax(axis=1),
target_names=lb.classes_))
print("CNN Accuracy: {:.2f}".format(
accuracy_score(testLabels.argmax(axis=1), cnnResults.argmax(axis=1))))
print("Cohen's Kappa {:.2f}".format(
cohen_kappa_score(testLabels.argmax(axis=1),
cnnResults.argmax(axis=1))))
input("")
# with open('../feature_names.pickle', 'r') as pickled:
# feature_names = pickle.load(pickled)
print "Loaded data; testing classifier..."
features_train, labels_train = ClassBalancingClassifierWrapper.rebalance(
features_train, labels_train, ratio=2)
results = []
for i in range(15):
print 'Round', i
classifier = DecisionTreeClassifier()
classifier = SKLPipeline([('feature_selection',
SelectPercentile(f_classif, 1)),
('classification', classifier)])
classifier.fit(features_train, labels_train)
labels_test_predicted = classifier.predict(features_test)
results.append(diff_binary_vectors(labels_test_predicted,
labels_test_gold))
# support = classifier.steps[0][1].get_support(True)
# print 'Selected', len(support), 'features:'
# for index in support:
# print ' ', feature_names[index]
print 'Results:'
print ClassificationMetrics.average(results, False)
# Visualize last round
'''
bestIndex = j
bestP = data["pred_" + top16tags[j]][i]
labelY.append(bestIndex)
bestPossibleY.append(bestP)
print("Label Y stat:")
labelYStat = defaultdict(lambda: 0)
for ly in labelY:
labelYStat[ly] = labelYStat[ly] + 1
for i in range(0, 16):
print("\tindex " + str(i) + ": " + str(labelYStat[i]))
model_to_show = DecisionTreeClassifier(random_state=42, max_depth=5)
model = DecisionTreeClassifier(random_state=42, max_depth=30)
# model = RandomForestClassifier(n_estimators=25, random_state=42)
model.fit(X, labelY)
model_to_show.fit(X, labelY)
tree.export_graphviz(model_to_show,
out_file=OUTPUT_TREE_FILE,
feature_names=columns,
label='none')
predY = model.predict(X)
print("Pred Y stat:")
predYStat = defaultdict(lambda: 0)
for py in predY:
predYStat[py] = predYStat[py] + 1
for i in range(0, 16):
print("\tindex " + str(i) + ": " + str(predYStat[i]))
from sklearn.metrics import f1_score
from sklearn.datasets import make_classification
from sklearn.tree.tree import DecisionTreeClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
X, Y = make_classification(
n_samples=10000,
n_features=1000,
n_informative=10)
trai_x, test_x, trai_y, test_y = train_test_split(X, Y, train_size=0.8)
#
clf = DecisionTreeClassifier()
model = clf.fit(trai_x, trai_y)
trai_yp = model.predict(trai_x)
test_yp = model.predict(test_x)
print(
model,
"F1, Trai:%.6f, Test:%.6f" %
(f1_score(trai_y, trai_yp), f1_score(test_y, test_yp))
)
#
clf = LogisticRegression()
model = clf.fit(trai_x, trai_y)
trai_yp = model.predict(trai_x)
test_yp = model.predict(test_x)
def DeepBoost(features,
labels,
n_steps,
max_depth_range,
PARAM_lambda,
PARAM_beta,
initial_weights=None):
num_features = features.shape[1]
sample_size = features.shape[0]
if (initial_weights == None):
weights = np.ones(sample_size) / sample_size
normalizer = np.exp(1) * sample_size
clf_list = []
for t in range(n_steps):
best_error = 0
best_grad = 0
best_index = -1 #?
old_tree_is_best = False
for j in range(len(clf_list)):
triple = clf_list[j]
alpha = triple[1]
if (abs(alpha) >= kTolerance):
old_clf = triple[0]
tree_depth = triple[2]
error = eval_clf(old_clf, features, labels, weights)
edge = error - 0.5
sign_edge = np.sign(edge)
grad = gradient(error, tree_depth, alpha, sign_edge,
sample_size, num_features, normalizer,
PARAM_lambda, PARAM_beta)
# print ('depth', tree_depth, 'error', error, 'grad', grad)
if (abs(grad) > abs(best_grad)):
best_grad = grad
best_error = error
best_index = j
old_tree_is_best = True
best_depth = -1
for depth in max_depth_range:
new_clf = DecisionTreeClassifier(max_depth=depth)
new_clf = new_clf.fit(features, labels, sample_weight=weights)
new_error = eval_clf(new_clf, features, labels, weights)
new_edge = new_error - 0.5
new_sign_edge = np.sign(new_edge)
new_grad = gradient(new_error, depth, 0, new_sign_edge,
sample_size, num_features, normalizer,
PARAM_lambda, PARAM_beta)
if (abs(new_grad) > abs(best_grad)):
best_new_clf = new_clf
best_grad = new_grad
best_error = new_error
best_depth = depth
old_tree_is_best = False
if old_tree_is_best:
triple = clf_list[best_index]
alpha = triple[1]
clf = triple[0]
depth = triple[2]
eta = compute_eta(best_error, depth, alpha, sample_size,
num_features, normalizer, PARAM_lambda,
PARAM_beta)
clf_list[best_index][1] += eta
else:
alpha = 0
clf = best_new_clf
depth = best_depth
#print ('t', t, 'best_error', best_error)
eta = compute_eta(best_error, depth, alpha, sample_size,
num_features, normalizer, PARAM_lambda,
PARAM_beta)
clf_list.append([clf, eta, depth])
old_normalizer = normalizer
normalizer = 0
y_predict = clf.predict(features)
for i in range(sample_size):
if (labels[i] == y_predict[i]):
u = eta
else:
u = -eta
weights[i] *= np.exp(-u)
normalizer += weights[i]
weights = weights / normalizer
normalizer = normalizer * old_normalizer
return clf_list, weights
print("rmse: " + str(rmse))
rmse = rmseEval(data["tw"]['target'], combinedPrediction2)[1]
print("rmse: " + str(rmse))
print("identification test:")
identificationColumns = []
for c in columns["all"]:
if c not in ['target', 'prediction', 'timestamp', 'location']:
identificationColumns.append(c)
clf = DecisionTreeClassifier()
traintestX = generateTrainingData(data["all"], identificationColumns)
clf = clf.fit(traintestX, label)
prediction = clf.predict(traintestX)
a = accuracy_score(label, prediction)
print(str(a))
a = accuracy_score(label, prediction, normalize=False)
print(str(a))
a = confusion_matrix(label, prediction)
print(str(a))
# with open(OUTPUT_DIRECTORY + "dt.dot", 'w') as f:
# f = tree.export_graphviz(clf, out_file=f, feature_names=identificationColumns)#, max_depth=10)
def train(dataset):
if (dataset == 'spambase'):
features, labels, testing_features, true_labels = fetch_data_from_raw(
'spambase')
else:
features, labels, testing_features, true_labels = fetch_npy_data(
dataset)
PARAM_lambda = 0.001
PARAM_lambda_2 = 0.01
PARAM_beta = 0.001
gamma = 0.06
tree_depth = 15
depth_range = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
## DeepBoost
T = 200
clf_list_db, weights = DeepBoost(features, labels, T, depth_range,
PARAM_lambda, PARAM_beta)
train_error_db = testEnsemble(clf_list_db, features, labels)
test_error_db = testEnsemble(clf_list_db, testing_features, true_labels)
print('db done')
## Deep BBM
gamma_list = [0.15, 0.1, 0.08, 0.06]
lambda_2_list = [1, 0.1, 0.01, 0.001, 0.0001]
max_depth_list = [1, 2, 3, 5, 10, 15]
# parameter search for Deep BBM
# train_errors_dbbm = np.zeros([len(gamma_list), len(lambda_2_list), len(max_depth_list)])
# test_errors_dbbm = np.zeros([len(gamma_list), len(lambda_2_list), len(max_depth_list)])
# for i in range(len(gamma_list)):
# for j in range(len(lambda_2_list)):
# for k in range(len(max_depth_list)):
# gamma = gamma_list[i]
# lambda_2 = lambda_2_list[j]
# max_depth = max_depth_list[k]
# depth_range = []
# for l in range(max_depth):
# depth_range.append(l+1)
# print (depth_range)
# clf_list_dbbm, weights = DeepBBM(features, labels, gamma, depth_range, lambda_2)
# train_error_dbbm = testEnsemble(clf_list_dbbm, features, labels)
# test_error_dbbm = testEnsemble(clf_list_dbbm, testing_features, true_labels)
# print ('ga', gamma, 'l2', lambda_2, 'md', max_depth, 'TrErr', train_error_dbbm, 'TeErr', test_error_dbbm)
# train_errors_dbbm[i, j, k] = train_error_dbbm
# test_errors_dbbm[i, j, k] = test_error_dbbm
# np.save('TrErr_dbbm_ps_2', train_errors_dbbm)
# np.save('TeErr_dbbm_ps_2', test_errors_dbbm)
clf_list_dbbm, weights = DeepBBM(features, labels, gamma, depth_range,
PARAM_lambda_2)
train_error_dbbm = testEnsemble(clf_list_dbbm, features, labels)
test_error_dbbm = testEnsemble(clf_list_dbbm, testing_features,
true_labels)
print('dbbm done')
## DecisionTreeClassifier
dtc = DecisionTreeClassifier(max_depth=tree_depth)
dtc = dtc.fit(features, labels)
train_pred = dtc.predict(features)
train_mse_dtc = ((train_pred - labels)**2).mean(axis=0)
test_pred = dtc.predict(testing_features)
# print (np.concatenate((np.expand_dims(pred, axis=1), np.expand_dims(true_labels, axis=1)), axis=1))
test_mse_dtc = ((test_pred - true_labels)**2).mean(axis=0)
## Boost by Majority
# gamma = 0.1
clf_list_bbm, weights = BoostByMaj(features, labels, tree_depth, gamma)
train_error_bbm = testEnsemble(clf_list_bbm, features, labels)
test_error_bbm = testEnsemble(clf_list_bbm, testing_features, true_labels)
#PlotMarginDistribution(clf_list_bbm, testing_features, true_labels)
print('bbm done')
# ## AdaBoost
# T = 200
clf_list_adb = AdaBoostClf(features, labels, tree_depth, T)
train_error_adb = testEnsemble(clf_list_adb, features, labels)
test_error_adb = testEnsemble(clf_list_adb, testing_features, true_labels)
print('adb done')
#PlotMarginDistribution(clf_list_adb, testing_features, true_labels)
## MarginBoost (from our homework)
# T = 200
margin = pow(2, -6)
clf_list_mb = MarginBoostClf(features, labels, tree_depth, T, margin)
train_error_mb = testEnsemble(clf_list_mb, features, labels)
test_error_mb = testEnsemble(clf_list_mb, testing_features, true_labels)
print('mb done')
#PlotMarginDistribution(clf_list_mgb, testing_features, true_labels)
## BrownBoost
total_time = 100
clf_list_brown = BrownBoost(features, labels, tree_depth, total_time)
train_error_brown = testEnsemble(clf_list_brown, features, labels)
test_error_brown = testEnsemble(clf_list_brown, testing_features,
true_labels)
print('bb done')
print('DeepBoost: train_error', train_error_db)
print('DeepBoost: test_error', test_error_db)
print('DeepBBM: train_error', train_error_dbbm)
print('DeepBBM: test_error', test_error_dbbm)
print('decision tree: train_mse', train_mse_dtc)
print('decision tree: test_mse', test_mse_dtc)
print('BBM: train_error', train_error_bbm)
print('BBM: test_error', test_error_bbm)
print('AdaBoost: train_error', train_error_adb)
print('AdaBoost: test_error', test_error_adb)
print('MarginBoost: train_error', train_error_mb)
print('MarginBoost: test_error', test_error_mb)
print('BrownBoost: train_error', train_error_brown)
print('BrownBoost: test_error', test_error_brown)
def decision_tree_fit(X,y):
clf = DecisionTreeClassifier(min_samples_leaf=5, random_state=42)
return clf.fit(X, y)
'''
Random forest
'''
print('Run random forest....')
rr_model = RandomForestClassifier(n_estimators=100,max_depth=10, random_state=1)
rr_model.fit(rel_train_X.relation_matrix, rel_train_Y.values)
rf_pred_train = rr_model.predict(rel_train_X.relation_matrix)
train_result.append(('RF', evaluateByF1(rf_pred_train, rel_train_Y.values)))
rf_pred_test = rr_model.predict(test_X)
test_result.append(('RF', evaluateByF1(rf_pred_test, test_Y)))
print('Run decision tree....')
id3_model = DecisionTreeClassifier(max_depth=10, random_state=1)
id3_model.fit(rel_train_X.relation_matrix, rel_train_Y.values)
id3_pred_train = id3_model.predict(rel_train_X.relation_matrix)
train_result.append(('ID3', evaluateByF1(id3_pred_train, rel_train_Y.values)))
id3_pred_test = id3_model.predict(test_X)
test_result.append(('ID3', evaluateByF1(id3_pred_test, test_Y)))
print('Performance of CBA and CMAR with different measures:')
printList(train_result)
printList(test_result)
from numpy import loadtxt
from sklearn.model_selection import train_test_split
from sklearn.tree.tree import DecisionTreeClassifier
func = lambda x: 0.0 if x == b'False' else 1.0
all_data = loadtxt('flare.csv',
delimiter=',',
skiprows=1,
converters={32: func})
target = all_data[:, -1]
data = all_data[:, 0:-1]
train_x, test_x, train_y, test_y = train_test_split(data,
target,
test_size=0.25,
random_state=100)
# SEIF A - Train the tree without trimming
clf = DecisionTreeClassifier(criterion="entropy")
clf.fit(train_x, train_y)
success_rate = clf.score(test_x, test_y)
# SEIF B - Trim to leaves smaller or equal to 20
clf = DecisionTreeClassifier(criterion="entropy", min_samples_leaf=20)
clf.fit(train_x, train_y)
success_rate = clf.score(test_x, test_y)
y_train_age = data[pd.notnull(data.Age)][['Age']]
regresor.fit(X_train_age, y_train_age)
# TODO: sprawdzić tą predykcję wieku, działa chyba ok
# data['AgePredicted'] = np.where(pd.isnull(data.Age), regresor.predict(data[['Title', 'SibSp', 'Parch']]), None)
data['Age'] = np.where(pd.isnull(data.Age),
regresor.predict(data[['Title', 'SibSp', 'Parch']]),
data['Age'])
##predykcja poziomu
classifier = DecisionTreeClassifier(max_depth=3, min_samples_leaf=2)
#
X_train_floor = data[pd.notnull(data.Floor)][['Embarked', 'Pclass']]
y_train_floor = data[pd.notnull(data.Floor)]['Floor'].values.astype('int')
classifier.fit(X_train_floor, y_train_floor)
data['Floor'] = np.where(pd.isnull(data.Floor),
classifier.predict(data[['Embarked', 'Pclass']]),
data['Floor'])
##zmiana ceny za bilet
data['TicketCounts'] = data.groupby('Ticket')['Ticket'].transform('count')
data['Fare'] = data['Fare'] / data['TicketCounts']
##usunięcie nieużywanych kolumn
data = data.drop(['Ticket', 'Cabin', 'Name', 'SibSp', 'Parch', 'TicketCounts'],
axis=1)
##zalozenie indeksu na kolumnie
store_pkl(audit_mapper, "Audit.pkl")
audit_X = audit[:, 0:48]
audit_y = audit[:, 48]
audit_y = audit_y.astype(int)
print(audit_X.dtype, audit_y.dtype)
def predict_audit(classifier):
adjusted = DataFrame(classifier.predict(audit_X), columns = ["Adjusted"])
adjusted_proba = DataFrame(classifier.predict_proba(audit_X), columns = ["probability_0", "probability_1"])
return pandas.concat((adjusted, adjusted_proba), axis = 1)
audit_tree = DecisionTreeClassifier(random_state = 13, min_samples_leaf = 5)
audit_tree.fit(audit_X, audit_y)
store_pkl(audit_tree, "DecisionTreeAudit.pkl")
store_csv(predict_audit(audit_tree), "DecisionTreeAudit.csv")
audit_forest = RandomForestClassifier(random_state = 13, min_samples_leaf = 5)
audit_forest.fit(audit_X, audit_y)
store_pkl(audit_forest, "RandomForestAudit.pkl")
store_csv(predict_audit(audit_forest), "RandomForestAudit.csv")
audit_regression = LogisticRegression()
audit_regression.fit(audit_X, audit_y)
store_pkl(audit_regression, "RegressionAudit.pkl")
store_csv(predict_audit(audit_regression), "RegressionAudit.csv")
def wrapper_for_decision_tree_in_sklearn(X, y, current_state_to_predict): clf = DecisionTreeClassifier() clf.fit(X, y) current_state_to_predict = np.array(current_state_to_predict).reshape(1,-1) predicted_state = clf.predict(current_state_to_predict) return predicted_state
dot_filename = mkstemp(suffix='.dot', dir=tmp_dir)[1]
with open(dot_filename, "w") as out_file:
export_graphviz(clf, out_file=out_file,
feature_names=feature_names,
class_names=class_names,
filled=True, rounded=True,
special_characters=True)
from IPython.display import Image
image_filename = image_filename or ('%s.png' % dot_filename)
subprocess.call(('dot -Tpng -o %s %s' %
(image_filename, dot_filename)).split(' '))
image = Image(filename=image_filename)
os.remove(dot_filename)
return image
from sklearn import datasets
from sklearn.tree.tree import DecisionTreeClassifier
%matplotlib inline
iris = datasets.load_iris()
X = iris.data
Y = iris.target
clf = DecisionTreeClassifier(max_depth=3)
clf.fit(X, Y)
convert_decision_tree_to_ipython_image(clf, image_filename='tree.png')
