def output_json(file="forest", size=100, prune=True, seed=50, burn=1):
random.seed(seed)
if file == "forest":
instances = ds.load_forest_fires()
variables = False
elif file == "voting":
instances = ds.load_congressional_voting()
variables = False
elif file == "iris":
instances = ds.load_iris()
variables = False
elif file == "mushroom":
instances = ds.load_mushroom()
variables = False
elif file == "rb_com_11":
instances = ds.load_rb_com_11()
variables = True
elif file == "rb_s_07":
instances = ds.load_rb_s_07()
variables = True
elif file == "rb_s_13":
instances = ds.load_rb_s_13()
variables = True
elif file == "rb_wb_03":
instances = ds.load_rb_wb_03()
variables = True
else:
instances = ds.load_forest_fires()
variables = False
random.shuffle(instances)
pprint.pprint(instances[0])
instances = instances[:size]
print(len(instances))
if variables:
variablizer = ObjectVariablizer()
instances = [variablizer.transform(t) for t in instances]
tree = TrestleTree()
tree.fit(instances, iterations=burn)
pprint.pprint(tree.root.output_json())
with open('output.js', 'w') as out:
out.write("var trestle_output = ")
out.write(json.dumps(tree.root.output_json()))
out.write(";")
def output_json(file="forest", size=100, prune=True, seed=50, burn=1):
random.seed(seed)
if file == "forest":
instances = ds.load_forest_fires()
variables = False
elif file == "voting":
instances = ds.load_congressional_voting()
variables = False
elif file == "iris":
instances = ds.load_iris()
variables = False
elif file == "mushroom":
instances = ds.load_mushroom()
variables = False
elif file == "rb_com_11":
instances = ds.load_rb_com_11()
variables = True
elif file == "rb_s_07":
instances = ds.load_rb_s_07()
variables = True
elif file == "rb_s_13":
instances = ds.load_rb_s_13()
variables = True
elif file == "rb_wb_03":
instances = ds.load_rb_wb_03()
variables = True
else:
instances = ds.load_forest_fires()
variables = False
random.shuffle(instances)
pprint.pprint(instances[0])
instances = instances[:size]
print(len(instances))
if variables:
variablizer = ObjectVariablizer()
instances = [variablizer.transform(t) for t in instances]
tree = TrestleTree()
tree.fit(instances, iterations=burn)
# pprint.pprint(tree.root.output_json())
with open('output.js', 'w') as out:
out.write("var trestle_output = ")
out.write(json.dumps(tree.root.output_json()))
out.write(";")
def calculate_aris(dataset):
shuffle(dataset)
dataset = dataset[:60]
variablizer = ObjectVariablizer()
dataset = [variablizer.transform(t) for t in dataset]
tree = TrestleTree()
tree.fit(dataset)
clusters = [cluster_split_search(tree, dataset, h, minsplit=1, maxsplit=40,
mod=False) for h in hueristics]
human_labels = [ds['_human_cluster_label'] for ds in dataset]
return [max(adjusted_rand_score(human_labels, huer), 0.01) for huer in
clusters]
class ScikitTrestle(object):
def __init__(self, params=None):
if params is None:
self.tree = TrestleTree()
else:
self.tree = TrestleTree(**params)
def ifit(self, x, y):
x = deepcopy(x)
x['_y_label'] = "%i" % y
self.tree.ifit(x)
def fit(self, X, y):
X = deepcopy(X)
for i, x in enumerate(X):
x['_y_label'] = "%i" % y[i]
self.tree.fit(X, randomize_first=False)
def predict(self, X):
return [int(self.tree.categorize(x).predict('_y_label')) for x in X]
class ScikitTrestle(object):
def __init__(self, **kwargs):
self.tree = TrestleTree(**kwargs)
self.state_format = "variablized_state"
def ifit(self, x, y):
x = deepcopy(x)
x['_y_label'] = float(y)
self.tree.ifit(x)
def fit(self, X, y):
X = deepcopy(X)
for i, x in enumerate(X):
x['_y_label'] = float(y)
self.tree.fit(X, randomize_first=False)
def skill_info(self, X):
raise NotImplementedError("Not implemented Erik H. says there is a way \
to serialize this -> TODO")
def predict(self, X):
return [self.tree.categorize(x).predict('_y_label') for x in X]
def calculate_aris(dataset):
shuffle(dataset)
dataset = dataset[:60]
variablizer = ObjectVariablizer()
dataset = [variablizer.transform(t) for t in dataset]
tree = TrestleTree()
tree.fit(dataset)
clusters = [
cluster_split_search(tree,
dataset,
h,
minsplit=1,
maxsplit=40,
mod=False) for h in hueristics
]
human_labels = [dataset['_human_cluster_label'] for dataset in dataset]
return [
max(adjusted_rand_score(human_labels, huer), 0.01) for huer in clusters
]
from concept_formation.visualize import visualize # These lines load up and use one of the example datasets included in the # library if you don't have a readily available dataset to test. The rb_s_07 # dataset is similar to but not exactly the same as the one used to generate # the figures in the paper. from concept_formation.datasets import load_rb_s_07 from concept_formation.preprocessor import ObjectVariablizer data = load_rb_s_07() # As long as your data conforms to the instance representation: # https://concept-formation.readthedocs.io/en/latest/instance_representation.html # it can be basically anything. # data = [] # This step is to make sure the component attributes of the instances are # properly tagged as variable. See the instance representation link above for # this. # ov = ObjectVariablizer() # data = ov.batch_transform(data) # These three lines are the core of the process. They will fit the data and # generate a visualization that will automatically open a browser to the view. # If you want to embed the output in some other process, like a LearnSphere # workflow, it would take a little more work but is easy in principle. tree = TrestleTree() tree.fit(data) visualize(tree, "vizfiles")
from concept_formation.cluster import cluster_split_search
from concept_formation.cluster import AIC, BIC, AICc, CU
from concept_formation.datasets import load_rb_wb_03
from concept_formation.preprocessor import ObjectVariablizer
seed(5)
towers = load_rb_wb_03()
shuffle(towers)
towers = towers[:60]
variablizer = ObjectVariablizer()
towers = [variablizer.transform(t) for t in towers]
tree = TrestleTree()
tree.fit(towers)
hueristics = [AIC, BIC, CU, AICc]
clusters = [
cluster_split_search(tree, towers, h, minsplit=1, maxsplit=40, mod=False)
for h in hueristics
]
human_labels = [tower['_human_cluster_label'] for tower in towers]
x = np.arange(len(hueristics))
y = [max(adjusted_rand_score(human_labels, huer), 0.01) for huer in clusters]
width = 0.45
hueristic_names = ['AIC', 'BIC', 'CU', 'AICc']
for i in range(len(clusters)):
seed(0)
X = np.sort(5 * rng.rand(80, 1), axis=0)
y = np.sin(X).ravel()
y[::5] += 3 * (0.5 - rng.rand(16))
# Fit regression models (Decision Tree and TRESTLE)
# For TRESTLE the y attribute is hidden, so only the X is used to make
# predictions.
dtree = DecisionTreeRegressor(max_depth=3)
dtree.fit(X, y)
ttree = TrestleTree()
training_data = [{
'x': float(X[i][0]),
'_y': float(y[i])
} for i, v in enumerate(X)]
ttree.fit(training_data, iterations=1)
# Predict
X_test = np.arange(0.0, 5.0, 0.01)[:, np.newaxis]
y_dtree = dtree.predict(X_test)
y_trestle = [ttree.categorize({'x': float(v)}).predict('_y') for v in X_test]
# Plot the results
plt.figure()
plt.scatter(X, y, c="k", label="Data")
plt.plot(X_test, y_trestle, c="g", label="TRESTLE", linewidth=2)
plt.plot(X_test, y_dtree, c="r", label="Decison Tree (Depth=3)", linewidth=2)
plt.xlabel("Data")
plt.ylabel("Target")
plt.title("TRESTLE/Decision Tree Regression")
plt.legend(loc=3)