Skip to content

Implementation of DBSCAN model #75

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 9 commits into
base: develop
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion Makefile
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ setup: ## Setup virtual environment for local development
&& $(MAKE) install-requirements

install-requirements:
pip install -U -e .
pip install --default-timeout=1000 -U -e .
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why increasing the timeout threshold ?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Timeout occurred while installing the packages because of network instability.


test: ## Run tests
python3 setup.py test
Expand Down
1 change: 1 addition & 0 deletions sqlflow_models/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@
from .rnn_based_time_series import RNNBasedTimeSeriesModel
from .auto_estimator import AutoClassifier, AutoRegressor
from .native_keras import RawDNNClassifier
from .dbscan import DBSCAN
try:
# NOTE: statsmodels have version conflict on PAI
from .arima_with_stl_decomposition import ARIMAWithSTLDecomposition
Expand Down
195 changes: 195 additions & 0 deletions sqlflow_models/dbscan.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,195 @@
#!usr/bin/env python
# -*- coding:utf-8 _*-

"""
__author__ : tiankelang
__email__ : [email protected]
__file_name__ : dbscan.py
__create_time__ : 2020/07/01

demo iris:
%%sqlflow
SELECT * FROM iris.train
TO TRAIN sqlflow_models.DBSCAN
WITH
model.min_samples=10,
model.eps=0.3
INTO sqlflow_models.my_dbscan_model;
"""
import tensorflow as tf
from scipy.spatial.distance import pdist, squareform
from sklearn.base import BaseEstimator, ClusterMixin
import pandas as pd
from sklearn import datasets, metrics
import numpy as np
from scipy.spatial import KDTree
from sklearn.datasets.samples_generator import make_blobs
from sklearn.preprocessing import StandardScaler
import six

def optimizer():
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems that it's not a NN model, should we remove this function?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, there is no need to set optimizer.

# SGD is just a placeholder to avoid panic on SQLFLow traning
return tf.keras.optimizers.SGD(lr=0.1, momentum=0.9)


def loss():
return None


def prepare_prediction_column(prediction):
"""Return the class label of highest probability."""
return prediction.argmax(axis=-1)


def purity_score(y_true, y_pred):
# compute contingency matrix
contingency_matrix = metrics.cluster.contingency_matrix(y_true, y_pred)
# return purity
return np.sum(np.amax(contingency_matrix, axis=0)) / np.sum(contingency_matrix)


class DBSCAN(tf.keras.Model):
def __init__(self,
eps: float = 0.5,
min_samples: int = 5,
has_label=False,
feature_columns=None):
'''
:param eps: Neighborhood distance
:param min_samples:
The minimum number of samples required to form a class cluster
'''
super(DBSCAN, self).__init__(name='DBSCAN')
self.eps = eps
self.min_samples = min_samples
self.core_sample_indices_ = list()
self.components_ = None
self.labels_ = None
self.has_label = has_label

def fit_predict(self, X):
n_samples = len(X)

kd_tree = KDTree(X) # build KDTree

density_arr = np.array([len(kd_tree.query_ball_point(x, self.eps)) for x in X]) # 密度数组

visited_arr = [False for _ in range(n_samples)] # Access tag array

k = -1 # init class
self.labels_ = np.array([-1 for _ in range(n_samples)])

for sample_idx in range(n_samples):
if visited_arr[sample_idx]: # Skip visited samples
continue

visited_arr[sample_idx] = True

# Skip noise samples and boundary samples
if density_arr[sample_idx] == 1 or density_arr[sample_idx] < self.min_samples:
continue

# core object
else:
# Find all the core objects in the neighborhood, including themselves
cores = [idx for idx in kd_tree.query_ball_point(X[sample_idx], self.eps) if
density_arr[idx] >= self.min_samples]
k += 1
self.labels_[sample_idx] = k
self.core_sample_indices_.append(sample_idx)

while cores:
cur_core = cores.pop(0)
if not visited_arr[cur_core]:
self.core_sample_indices_.append(cur_core)
visited_arr[cur_core] = True
self.labels_[cur_core] = k

neighbors = kd_tree.query_ball_point(X[cur_core], self.eps)
neighbor_cores = [idx for idx in neighbors if
idx not in cores and density_arr[idx] >= self.min_samples]
neighbor_boards = [idx for idx in neighbors if density_arr[idx] < self.min_samples]

cores.extend(neighbor_cores)

for idx in neighbor_boards:
if self.labels_[idx] == -1:
self.labels_[idx] = k

# Update class properties
self.core_sample_indices_ = np.sort(np.array(self.core_sample_indices_))
self.components_ = X[self.core_sample_indices_.astype('int64')]
return self.labels_


def _read_Dataset_data(self, dataset):
data = None
label = None
flag = True
print("dataset:", dataset)
for item in dataset:
# print("item:", item)
if flag:
flag = False
item_data = item[0] # dict
len1 = len(item_data)
index=0

feature_data = []
feature_column_names = []

for k, v in item_data.items():
if index == (len1-1):
item_label = v.numpy().reshape(1, )
else:
feature_column_names.append(k)
feature_data.append(v.numpy())
index = index + 1
feature_data = np.asarray(feature_data).reshape(1, -1)

data = np.asarray(feature_data).reshape(1, -1)
label = item_label
else:
item_data = item[0]
len1 = len(item_data)
index = 0

feature_data = []
feature_column_names = []

for k, v in item_data.items():
if index == (len1 - 1):
item_label = v.numpy().reshape(1, )
else:
feature_column_names.append(k)
feature_data.append(v.numpy())
index = index + 1
feature_data = np.asarray(feature_data).reshape(1, -1)

data = np.concatenate((data, feature_data), axis=0)
label = np.concatenate((label, item_label), axis=0)
print("data:", type(data), data.shape)
print("label:", type(label), label.shape)
return data, label
# do custom training here, parameter "dataset" is a tf.dataset type representing the input data.
def sqlflow_train_loop(self, dataset, epochs=1, verbose=0):
'''
Parameter `epochs` and `verbose` will not be used in this function. :param dataset: demo iris,
:param dataset:
demo iris <class 'tensorflow.python.data.ops.dataset_ops.DatasetV1Adapter'>
<DatasetV1Adapter shapes: ({sepal_length: (1,), sepal_width: (1,), petal_length: (1,), petal_width: (1,)},
(1, None)), types: ({sepal_length: tf.float32, sepal_width: tf.float32, petal_length: tf.float32,
petal_width: tf.float32}, tf.int64)>
:param epochs:
:param verbose:
:return:
'''
data, label = self._read_Dataset_data(dataset)

self.fit_predict(data)
print("DBSCAN(eps= %.2f, minpts= %d), the purity score: %f" %
(self.eps,
self.min_samples,
purity_score(label, self.labels_)))
# print("Predict labels:", self.labels_)
# print("True labels:", label)
46 changes: 46 additions & 0 deletions tests/test_db_scan.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
import sqlflow_models
from tests.base import BaseTestCases
import tensorflow as tf
import unittest
import numpy as np
from sklearn.datasets import load_iris
from sklearn.datasets.samples_generator import make_blobs
from sklearn.preprocessing import StandardScaler
from sklearn import datasets, metrics
import logging
from pathlib import Path
from numpy import ndarray, testing

iris = datasets.load_iris()
iris_data = np.array(iris.data)
iris_target = iris.target


def purity_score(y_true, y_pred):
# compute contingency matrix
contingency_matrix = metrics.cluster.contingency_matrix(y_true, y_pred)
# return purity
return np.sum(np.amax(contingency_matrix, axis=0)) / np.sum(contingency_matrix)


def print_in_test(string):
logging.warning(string)


class TestDBSCAN(unittest.TestCase):
"""DBSCAN test cases."""

@classmethod
def setUpClass(self):
self.dbscan = sqlflow_models.DBSCAN(
min_samples=10, eps=.4)
self.dbscan.sqlflow_train_loop(iris_data)

def test_dbscan_return_labels_with_type_numpy_array(self):
self.assertIsInstance(self.dbscan.labels_, ndarray)
print("Test DBSCAN (minpts=10, eps=0.4), the purity score: %f" %
purity_score(iris_target, self.dbscan.labels_))


if __name__ == '__main__':
unittest.main()