def describe_model(model, X, Y, log, plot_all):
predictor_names = X.columns.values.tolist()
log += print_and_log("Model:")
log += print_and_log(model)
log += print_and_log("Predictor variables:")
for p in predictor_names:
log += print_and_log("\t{}".format(p))
# Export tree SVG
if plot_all:
from dtreeviz.trees import dtreeviz
log += print_and_log("\nExport tree to SVG:")
viz = dtreeviz(
model,
X.values,
Y.values.ravel(),
target_name="perf",
feature_names=predictor_names,
)
viz.save("trytree.svg")
viz.view()
return log
def viz_digits(orientation="TD",
max_depth=3,
random_state=666,
fancy=True,
pickX=False):
clf = tree.DecisionTreeClassifier(max_depth=max_depth,
random_state=random_state)
digits = load_digits()
# "8x8 image of integer pixels in the range 0..16."
columns = [f'pixel[{i},{j}]' for i in range(8) for j in range(8)]
clf.fit(digits.data, digits.target)
X = None
if pickX:
X = digits.data[np.random.randint(0, len(digits.data)), :]
viz = dtreeviz(clf,
digits.data,
digits.target,
target_name='number',
feature_names=columns,
orientation=orientation,
class_names=[chr(c) for c in range(ord('0'),
ord('9') + 1)],
fancy=fancy,
histtype='bar',
X=X)
return viz
def viz_tree(clf,Xdata,Ydata,label,outfile,point=None):
from dtreeviz.trees import dtreeviz
from sklearn.tree import export_graphviz
X = Xdata.pd
Y = Ydata.pd
X_headers = X.columns
Y_headers = Y.columns
# If "point" given as argument, find the closest datapoint to "point", will do treewalk for this observation
if(point is not None):
points = Xdata.vtk.points
dist = points-point
dist = np.sqrt(dist[:,0]**2.0 + dist[:,1]**2.0 + dist[:,2]**2.0)
loc = np.argmin(dist)
print('point = ', point)
print('nearest point = ',points[loc,:])
print('distance = ',dist[loc])
datapoint = X.iloc[loc]
else:
datapoint = None
# Extract the classifier object from the clf multilearn object
index = Y_headers.to_list().index(label)
clf = clf.classifiers_[index]
# TODO: check if clf is a decision tree
viz = dtreeviz(clf, X, Y[label],
feature_names=X_headers,
target_name=label,class_names=["False","True"],X=datapoint)
viz.save(outfile)
def viz_boston(orientation="TD", max_depth=3, random_state=666, fancy=True):
regr = tree.DecisionTreeRegressor(max_depth=max_depth,
random_state=random_state)
boston = load_boston()
regr = regr.fit(boston.data, boston.target)
X = boston.data[np.random.randint(0, len(boston.data)), :]
print(boston.feature_names)
features = np.array([
'CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX',
'PTRATIO', 'B', 'LSTAT'
])
viz = dtreeviz(regr,
boston.data,
boston.target,
target_name='price',
feature_names=features,
orientation=orientation,
fancy=fancy,
X=X)
export_graphviz(regr,
out_file="/tmp/boston-scikit-tree.dot",
filled=True,
rounded=True,
special_characters=True)
return viz
def viz_diabetes(orientation="TD",
max_depth=3,
random_state=666,
fancy=True,
pickX=False):
diabetes = load_diabetes()
regr = tree.DecisionTreeRegressor(max_depth=max_depth,
random_state=random_state)
regr.fit(diabetes.data, diabetes.target)
X = None
if pickX:
X = diabetes.data[np.random.randint(0, len(diabetes.data)), :]
viz = dtreeviz(regr,
diabetes.data,
diabetes.target,
target_name='progr',
feature_names=diabetes.feature_names,
orientation=orientation,
fancy=fancy,
X=X)
return viz
def viz_knowledge(orientation="TD", max_depth=3, random_state=666, fancy=True):
# data from https://archive.ics.uci.edu/ml/datasets/User+Knowledge+Modeling
clf = tree.DecisionTreeClassifier(max_depth=max_depth,
random_state=random_state)
know = pd.read_csv("data/knowledge.csv")
target_names = ['very_low', 'Low', 'Middle', 'High']
know['UNS'] = know['UNS'].map({n: i for i, n in enumerate(target_names)})
X_train, y_train = know.drop('UNS', axis=1), know['UNS']
clf = clf.fit(X_train[['PEG', 'LPR']], y_train)
X = X_train.iloc[np.random.randint(0, len(know))]
viz = dtreeviz(
clf,
X_train[['PEG', 'LPR']],
y_train,
target_name='UNS',
feature_names=['PEG', 'LPR'],
orientation=orientation,
class_names=target_names,
# show_node_labels=True,
histtype='strip',
fancy=fancy)
return viz
def visualize(clf):
(train, target) = read_train_data()
viz = dtreeviz(clf, train, target, target_name='HSV', feature_names=['Hue', 'Saturation', 'Value'],
class_names=['Yellow', 'Green',
'Red', 'Blue', 'Black', 'White']
)
viz.view()
def export_tree_advanced(self,X,Y,feature_names,filename_out):
viz = dtreeviz(self.model, X, Y, feature_names=feature_names,colors={'classes': [None, None,['#0080FF80', '#FF800080']]})
viz.save(filename_out)
ext = filename_out.split('.')[-1]
name = filename_out.split('.' + ext)[0]
if os.path.isfile(name):
os.remove(name)
return
# ----------------------------------------------------------------------------------------------------------------------
def interpret(
self,
X_train,
y_train,
X_validation,
y_validation,
model_file_path,
learner_name,
target_name=None,
class_names=None,
metric_name=None,
ml_task=None,
explain_level=2,
):
super(DecisionTreeAlgorithm, self).interpret(
X_train,
y_train,
X_validation,
y_validation,
model_file_path,
learner_name,
target_name,
class_names,
metric_name,
ml_task,
explain_level,
)
if explain_level == 0:
return
with warnings.catch_warnings():
warnings.simplefilter(action="ignore")
try:
if len(class_names) > 10:
# dtreeviz does not support more than 10 classes
return
viz = dtreeviz(
self.model,
X_train,
y_train,
target_name="target",
feature_names=X_train.columns,
class_names=class_names,
)
tree_file_plot = os.path.join(model_file_path,
learner_name + "_tree.svg")
viz.save(tree_file_plot)
except Exception as e:
logger.info(
f"Problem when visualizing decision tree. {str(e)}")
save_rules(self.model, X_train.columns, class_names,
model_file_path, learner_name)
def explain(skater_exp: Explanation, training_df, test_df, explanation_target,
prefix_target):
job = skater_exp.job
model = joblib.load(job.predictive_model.model_path)
model = model[0]
features = list(training_df.drop(['trace_id', 'label'], 1).columns.values)
interpreter = Interpretation(training_df, feature_names=features)
X_train = training_df.drop(['trace_id', 'label'], 1)
Y_train = training_df['label'].values
model_inst = InMemoryModel(model.predict,
examples=X_train,
model_type=model._estimator_type,
unique_values=[1, 2],
feature_names=features,
target_names=['label'])
surrogate_explainer = interpreter.tree_surrogate(model_inst, seed=5)
surrogate_explainer.fit(X_train,
Y_train,
use_oracle=True,
prune='post',
scorer_type='default')
surrogate_explainer.class_names = features
viz = dtreeviz(surrogate_explainer.estimator_,
X_train,
Y_train,
target_name='label',
feature_names=features,
orientation="TD",
class_names=list(surrogate_explainer.class_names),
fancy=True,
X=None,
label_fontsize=12,
ticks_fontsize=8,
fontname="Arial")
name = create_unique_name("skater_plot.svg")
viz.save(name)
if os.path.getsize(name) > 15000000:
return 'The file size is too big'
f = open(name, "r")
response = f.read()
os.remove(name)
if os.path.isfile(name.split('.svg')[0]):
os.remove(name.split('.svg')[0])
return response
def viz_iris(orientation="TD", max_depth=5, random_state=666, fancy=True):
clf = tree.DecisionTreeClassifier(max_depth=max_depth,
random_state=random_state)
iris = load_iris()
data = pd.DataFrame(iris.data)
data.columns = iris.feature_names
clf = clf.fit(data, iris.target)
# for i in range(len(iris.data)):
for i in [60]:
x = data.iloc[i]
pred = clf.predict([x.values])
shadow_tree = ShadowDecTree(
clf,
iris.data,
iris.target,
feature_names=iris.feature_names,
class_names=["setosa", "versicolor", "virginica"])
pred2 = shadow_tree.predict(x.values)
print(
f'{x} -> {pred[0]} vs mine {pred2[0]}, path = {[f"node{p.feature_name()}" for p in pred2[1]]}'
)
path = [n.id for n in pred2[1]]
if pred[0] != pred2[0]:
print("MISMATCH!")
features = list(data.columns)
features = np.array([
'sepal length (cm)', 'sepal width (cm)', 'petal length (cm)',
'petal width (cm)'
])
st = dtreeviz(
clf,
iris.data,
iris.target,
target_name='variety',
feature_names=features,
orientation=orientation,
class_names=["setosa", "versicolor", "virginica"], # 0,1,2 targets
#histtype='strip',
fancy=fancy,
X=x)
return st
def simple_dtree(df,
x_list,
y_var,
max_depth=3,
regressor=False,
min_samples_split=2,
test_size=0.3):
"""
instant DecisionTree model
"""
from sklearn.tree import DecisionTreeRegressor
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from dtreeviz.trees import dtreeviz
X = df[x_list]
y = df[y_var]
X_train, X_test, y_train, y_test = train_test_split(X,
y,
random_state=123,
test_size=test_size)
# build model, and fitting
if regressor:
model = DecisionTreeRegressor(max_depth=max_depth,
random_state=123,
min_samples_split=min_samples_split)
else:
model = DecisionTreeClassifier(max_depth=max_depth,
random_state=123,
min_samples_split=min_samples_split)
model.fit(X_train, y_train)
# plot
print('Train score: {:.3f}'.format(model.score(X_train, y_train)))
print('Test score: {:.3f}'.format(model.score(X_test, y_test)))
viz = dtreeviz(
model,
X_train=X,
y_train=y,
target_name=y_var,
feature_names=x_list,
precision=2,
class_names=None
if model.classes_ is None else model.classes_.tolist(),
)
viz.view()
return model
def weird_binary_case():
# See bug https://github.com/parrt/dtreeviz/issues/17
import numpy as np
from sklearn.tree import DecisionTreeClassifier
from dtreeviz.trees import dtreeviz
x = np.random.choice([-1, 1], size=(100, 2))
y = np.random.choice([0, 1], size=100)
viz = dtreeviz(tree_model=DecisionTreeClassifier(max_depth=1).fit(x, y),
X_train=x,
y_train=y,
feature_names=['a', 'b'],
target_name='y',
class_names=[1, 0])
return viz
def interpret(
self,
X_train,
y_train,
X_validation,
y_validation,
model_file_path,
learner_name,
target_name=None,
class_names=None,
metric_name=None,
ml_task=None,
explain_level=2,
):
super(DecisionTreeRegressorAlgorithm, self).interpret(
X_train,
y_train,
X_validation,
y_validation,
model_file_path,
learner_name,
target_name,
class_names,
metric_name,
ml_task,
explain_level,
)
if explain_level == 0:
return
try:
viz = dtreeviz(
self.model,
X_train,
y_train,
target_name="target",
feature_names=X_train.columns,
)
tree_file_plot = os.path.join(model_file_path,
learner_name + "_tree.svg")
viz.save(tree_file_plot)
except Exception as e:
logger.info(
f"Problem when visuzalizin decision tree regressor. {str(e)}")
save_rules(self.model, X_train.columns, None, model_file_path,
learner_name)
def explore_prediction():
"""Interfaz interactiva para ver como se hace una predicción al azar."""
x_sample = df.sample()
display(x_sample)
viz = dtreeviz.dtreeviz(
clf,
X,
y,
target_name='smoker',
feature_names=list(X.columns),
class_names=list(y_encoder.classes_),
scale=1.0,
X=x_sample[X.columns].iloc[0].values,
)
display(viz)
def interpret(self,
X,
y,
model_file_path,
learner_name,
target_name=None,
class_names=None):
try:
viz = dtreeviz(self.model,
X,
y,
target_name="target",
feature_names=X.columns)
self._tree_file_plot = os.path.join(model_file_path,
learner_name + "_tree.svg")
viz.save(self._tree_file_plot)
except Exception as e:
self._tree_file_plot = None
def visualize_decision_tree(self,
X_train,
Y_train,
feature_names,
max_depth=3):
X_train = self.sample_like(X_train, frac=self.sample_frac)
Y_train = self.sample_like(Y_train, frac=self.sample_frac)
Y_train = self.ravel_like(Y_train)
clf = DecisionTreeClassifier(max_depth=max_depth)
clf.fit(X_train, Y_train)
for pred_col in self.pred_cols:
viz = dtreeviz(clf,
X_train,
Y_train,
target_name=pred_col,
feature_names=feature_names,
class_names=list(set([str(i) for i in Y_train])))
display(viz)
def getDtreeVizImg(rand_state=42, max_depth=3):
fname = BASE_IMG_PATH + "dtv" + str(rand_state) + '_' + str(
max_depth) + '.png'
if (not os.path.exists(fname)):
X = df.drop(['Class'], axis=1)
y = df['Class']
# Split the training and test dataset.
# Since in this program the focus is not on predictions
# but on creating a new style decision tree, the test set is unused
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=rand_state)
# initialise the decision tree model
DecsTreeModel = DecisionTreeClassifier(criterion='entropy',
max_depth=max_depth)
# train the model with the training set
DecsTreeModel.fit(X_train, y_train)
# Creating the dtreeviz style decision tree
viz = dtreeviz(DecsTreeModel,
X_train,
y_train,
feature_names=list(X.columns.values),
class_names=['0', '1'])
# save the decision tree image in SVG format
viz.save(BASE_IMG_PATH + "svgtempfile.svg")
out = BytesIO()
#convert the decision tree image from SVG to PNG format
cairosvg.svg2png(url=BASE_IMG_PATH + "svgtempfile.svg", write_to=out)
dtree = Image.open(out)
dtree.save(fname)
with open(fname, "rb") as image:
img = image.read()
# The Base 64 format is useful for serialization of image data.
# Doing this enables the image data to be saved in peristent storage or transfer it over network.
barray = base64.b64encode(img).decode('utf-8')
return barray
def run_single_tree(X_train, y_train, X_test, y_test, depth, plot_tree,
feature_cols):
model = tree.DecisionTreeRegressor(max_depth=depth).fit(X_train, y_train)
accuracy_train = model.score(X_train, y_train)
accuracy_test = model.score(X_test, y_test)
y_predictions = model.predict(X_test)
y_predictions_series = pandas.Series(y_predictions, index=y_test.index)
def resid(row):
return row["true"] - row["pred"]
resid_df = pandas.concat([y_test, y_predictions_series],
keys=["true", "pred"],
axis=1)
resid_df["residual"] = resid_df.apply(resid, axis=1)
plt.scatter(resid_df["true"][:round(len(resid_df) / 30)],
resid_df["residual"][:round(len(resid_df) / 30)],
alpha=0.5)
plt.savefig(f"plots/residplot-depth-{depth}.png")
plt.clf()
print('Single tree depth: ', depth)
print('Accuracy, Training Set: ', round(accuracy_train * 100, 5), '%')
print('Accuracy, Test Set: ', round(accuracy_test * 100, 5), '%')
print('RMSE Tree, Test Set: ',
sqrt(mean_squared_error(y_test, y_predictions)))
if plot_tree:
viz = dtreeviz(model,
X_train,
y_train,
fancy=False,
target_name='chance',
feature_names=feature_cols)
viz.save("plots/plot.svg")
return accuracy_train, accuracy_test
def produce_tree_visualization(tree, tree_index, x, y, target_name,
feature_names, class_names, model_uid):
"""
Produces visualization of a decision tree from an ensemble.
:param tree: tree model
:param tree_index: index of the tree in the ensemble
:param x: predictor matrix
:param y: target series
:param target_name: name of the target
:param feature_names: list of feature names
:param class_names: name of the target classes
:param model_uid: name of the model
"""
viz = dtreeviz(tree.estimators_[tree_index],
x,
y,
target_name=target_name,
feature_names=feature_names,
class_names=class_names)
viz.save(
os.path.join('modeling', model_uid, 'diagnostics', 'trees',
f'decision_tree_{tree_index}.svg'))
def create_tree(df, max_depth, target='admitted', **kwargs):
X = df.drop(columns=[target])
y = df[target]
tree = DecisionTreeClassifier(
max_depth=max_depth,
**kwargs,
)
tree.fit(X, y)
rep = report(y, tree.predict(X))
viz = dtreeviz(
tree,
X,
y,
target_name='admitted',
feature_names=X.columns,
class_names=['no', 'yes'],
)
return tree, rep, viz
def describe_hpo(gs, X, Y, log, plot_all):
predictor_names = X.columns.values.tolist()
log += print_and_log(
"\n----------------------------------------------------------------------------"
)
log += print_and_log("Available predictor variables:")
for p in predictor_names:
log += print_and_log("\t{}".format(p))
log += print_and_log("\nBest parameters set found on development set:")
for bestpar_name, bestpar_value in gs.best_params_.items():
log += print_and_log("\t{}: {}".format(bestpar_name, bestpar_value))
log += print_and_log("\nBest estimator:")
best_estimator = gs.best_estimator_._final_estimator
log += print_and_log(best_estimator)
log += print_and_log(
"----------------------------------------------------------------------------"
)
# Export tree SVG
if plot_all:
from dtreeviz.trees import dtreeviz
log += print_and_log("\nExport tree to SVG:")
viz = dtreeviz(
best_estimator,
X.values,
Y.values.ravel(),
target_name="perf",
feature_names=predictor_names,
)
viz.save("trytree.svg")
viz.view()
return log
def viz_wine(orientation="TD",
max_depth=3,
random_state=666,
fancy=True,
pickX=False):
clf = tree.DecisionTreeClassifier(max_depth=max_depth)
wine = load_wine()
X_train = wine.data
y_train = wine.target
clf.fit(X_train, y_train)
X = None
if pickX:
X = X_train[np.random.randint(0, len(X_train.data)), :]
viz = dtreeviz(clf,
wine.data,
wine.target,
target_name='wine',
feature_names=wine.feature_names,
class_names=list(wine.target_names),
X=X) # pass the test observation
return viz
def interpret(self,
X,
y,
model_file_path,
learner_name,
target_name=None,
class_names=None):
try:
if len(class_names) > 10:
# dtreeviz does not support more than 10 classes
return
viz = dtreeviz(
self.model,
X,
y,
target_name="target",
feature_names=X.columns,
class_names=class_names,
)
self._tree_file_plot = os.path.join(model_file_path,
learner_name + "_tree.svg")
viz.save(self._tree_file_plot)
except Exception as e:
self._tree_file_plot = None
from dtreeviz.trees import dtreeviz
from sklearn import tree
from sklearn.datasets import load_wine
wine = load_wine()
classifier = tree.DecisionTreeClassifier(max_depth=2)
classifier.fit(wine.data, wine.target)
vis = dtreeviz(
classifier,
wine.data,
wine.target,
target_name="wine_type",
feature_names=wine.feature_names,
)
vis.view()
# .set_axis(['X0', 'X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10', 'X11',
# 'X12', 'X13', 'X14', 'X15', 'X16', 'X17', 'X18', 'X19', 'X20',
# 'X21', 'X22', 'X23', 'Y'], axis=1)\
# .to_csv('csv/accomodation_data.csv')
# 4 決定木によるモデル構築 --------------------------------------------------------
# モデル構築
# --- 決定木による分類器
clf = DecisionTreeClassifier(max_depth=2)
clf = clf.fit(X=features, y=data_o)
# 確認
vars(clf)
# 5 決定木の可視化 ---------------------------------------------------------------
# indexの抽出
time_index = df_info.resample('M').count().index
# 決定木を描画
viz = dtreeviz(
clf,
features,
data_o,
target_name='Class',
feature_names=time_index,
class_names=['False', 'True'],
)
viz
#print representation
text_representation = tree.export_text(clf)
print(text_representation)
#save to a file
with open('decisiontree1.log','w') as fout : fout.write(text_representation)
#plottree
tree.plot_tree(decision_tree=clf)
fig = plt.figure(figsize=(10,8))
_ = tree.plot_tree(clf, feature_names= iris.feature_names, class_names=iris.target_names, filled=True) #see plot
import graphviz
dot_data = tree.export_graphviz(decision_tree=clf, out_file=None, feature_names = iris.feature_names, class_names = iris.target_names, filled=True)
graph = graphviz.Source(dot_data, format='png')
#error dot path
import os
os.environ["PATH"] += os.pathsep + 'c:/Program Files (x86)/Graphviz2.38/bin/'
graph = graphviz.Source(dot_data, format='png')
graph
graph.render('DecisionTree.png ')
#%%%Plot DT with dtreeviz
#pip install dtreeviz
from dtreeviz.trees import dtreeviz
viz = dtreeviz(clf, X,y, target_name='target', feature_names = iris.feature_names, class_names = list(iris.target_names))
viz
viz.save('dt2.svg')
import pandas as pd
from sys import argv
from dtreeviz.trees import dtreeviz
from sklearn.metrics import f1_score
from sklearn.tree import DecisionTreeClassifier
data = pd.read_csv(argv[1])
cols = list(data.columns)
full = cols[1:-4] # the first one is the date and the last four are the labels
labels = cols[-4] # these are the raw labels (0, 1, 2)
names = []
for f in full:
if f in argv or len(argv) < 3:
names.append(f)
X = data[names]
y = data[labels]
print(names)
dt = DecisionTreeClassifier()
model = dt.fit(X, y)
v = dtreeviz(model,
X,
y,
target_name='trend',
feature_names=names,
class_names=['decrease', 'stable', 'increase'])
v.save('dt.svg')
print(f1_score(y, dt.predict(X), average='weighted'))
#正答率を求める
pre1 = clf_result.predict(X_train)
ac_score1 = metrics.accuracy_score(y_train, pre1)
print("トレーニングデータ正答率 = ", ac_score1)
pre2 = clf_result.predict(X_test)
ac_score2 = metrics.accuracy_score(y_test, pre2)
print("テストデータ正答率 = ", ac_score2)
#重要度の可視化
features = []
for s in range(len(X.columns)):
features.append(X.columns[s])
n_features = X.shape[1]
plt.barh(range(n_features), clf_result.feature_importances_, align="center")
plt.yticks(np.arange(n_features), features)
plt.xlabel("importance")
plt.ylabel("Feature value")
plt.savefig("barh.png")
#決定木の可視化
viz = dtreeviz(clf_result,
X.values,
y.values,
target_name="variety",
feature_names=features,
class_names=["0", "1", "2", "3"])
viz.view()
# DOT data
dot_data = tree.export_graphviz(clf, out_file=None,
feature_names=iris.feature_names,
class_names=iris.target_names,
filled=True)
# Draw graph
graph = graphviz.Source(dot_data, format="png")
graph
#graph.render("decision_tree_graphivz")
# Plot Decision Tree with dtreeviz Package
from dtreeviz.trees import dtreeviz # remember to load the package
viz = dtreeviz(clf, X, y,
target_name="target",
feature_names=iris.feature_names,
class_names=list(iris.target_names))
viz
################## REGRESSION TASK #######################
# Visualizing the Decision Tree in Regression Task
from sklearn import datasets
from sklearn.tree import DecisionTreeRegressor
from sklearn import tree
# Prepare the data data
boston = datasets.load_boston()
X = boston.data
