Most of the traditional clustering techniques, such as k-means, hierarchical and fuzzy clustering, can be used to group data without supervision.
However, when applied to tasks with arbitrary shape clusters, or clusters within cluster, the traditional techniques might be unable to achieve good results. That is, elements in the same cluster might not share enough similarity or the performance may be poor. Additionally, Density-based Clustering locates regions of high density that are separated from one another by regions of low density. Density, in this context, is defined as the number of points within a specified radius.
In this section, the main focus will be manipulating the data and properties of DBSCAN and observing the resulting clustering.
Table of contents
- Clustering with Randomly Generated Data
- Weather Station Clustering with DBSCAN & scikit-learn
Import the following libraries:
- numpy as np
- DBSCAN from sklearn.cluster
- make_blobs from sklearn.datasets.samples_generator
- StandardScaler from sklearn.preprocessing
- matplotlib.pyplot as plt
Remember %matplotlib inline to display plots
# Notice: For visualization of map, you need basemap package.
# if you dont have basemap install on your machine, you can use the following line to install it
# !conda install -c conda-forge basemap==1.1.0 matplotlib==2.2.2 -y
# Notice: you maight have to refresh your page and re-run the notebook after installation
import numpy as np
from sklearn.cluster import DBSCAN
from sklearn.datasets.samples_generator import make_blobs
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
%matplotlib inline
Data generation
The function below will generate the data points and requires these inputs:- centroidLocation: Coordinates of the centroids that will generate the random data.
- Example: input: [[4,3], [2,-1], [-1,4]]
- numSamples: The number of data points we want generated, split over the number of centroids (# of centroids defined in centroidLocation)
- Example: 1500
- clusterDeviation: The standard deviation between the clusters. The larger the number, the further the spacing.
- Example: 0.5
def createDataPoints(centroidLocation, numSamples, clusterDeviation):
# Create random data and store in feature matrix X and response vector y.
X, y = make_blobs(n_samples=numSamples, centers=centroidLocation,
cluster_std=clusterDeviation)
# Standardize features by removing the mean and scaling to unit variance
X = StandardScaler().fit_transform(X)
return X, y
Use createDataPoints with the 3 inputs and store the output into variables X and y.
X, y = createDataPoints([[4,3], [2,-1], [-1,4]] , 1500, 0.5)
Modeling
DBSCAN stands for Density-Based Spatial Clustering of Applications with Noise. This technique is one of the most common clustering algorithms which works based on density of object. The whole idea is that if a particular point belongs to a cluster, it should be near to lots of other points in that cluster.It works based on two parameters: Epsilon and Minimum Points
Epsilon determine a specified radius that if includes enough number of points within, we call it dense area
minimumSamples determine the minimum number of data points we want in a neighborhood to define a cluster.
epsilon = 0.3
minimumSamples = 7
db = DBSCAN(eps=epsilon, min_samples=minimumSamples).fit(X)
labels = db.labels_
labels
Distinguishing Outliers
Lets Replace all elements with 'True' in core_samples_mask that are in the cluster, 'False' if the points are outliers.# First, create an array of booleans using the labels from db.
core_samples_mask = np.zeros_like(db.labels_, dtype=bool)
core_samples_mask[db.core_sample_indices_] = True
core_samples_mask
# Number of clusters in labels, ignoring noise if present.
n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
n_clusters_
# Remove repetition in labels by turning it into a set.
unique_labels = set(labels)
unique_labels
Data visualization
# Create colors for the clusters.
colors = plt.cm.Spectral(np.linspace(0, 1, len(unique_labels)))
colors
# Plot the points with colors
for k, col in zip(unique_labels, colors):
if k == -1:
# Black used for noise.
col = 'k'
class_member_mask = (labels == k)
# Plot the datapoints that are clustered
xy = X[class_member_mask & core_samples_mask]
plt.scatter(xy[:, 0], xy[:, 1],s=50, c=col, marker=u'o', alpha=0.5)
# Plot the outliers
xy = X[class_member_mask & ~core_samples_mask]
plt.scatter(xy[:, 0], xy[:, 1],s=50, c=col, marker=u'o', alpha=0.5)
Practice¶
To better underestand differences between partitional and density-based clusteitng, try to cluster the above dataset into 3 clusters using k-Means.
Notice: do not generate data again, use the same dataset as above.
# write your code here
from sklearn.cluster import KMeans
k = 3
k_means3 = KMeans(init = "k-means++", n_clusters = k, n_init = 12)
k_means3.fit(X)
fig = plt.figure(figsize=(6, 4))
ax = fig.add_subplot(1, 1, 1)
for k, col in zip(range(k), colors):
my_members = (k_means3.labels_ == k)
plt.scatter(X[my_members, 0], X[my_members, 1], c=col, marker=u'o', alpha=0.5)
plt.show()
Double-click here for the solution.
Weather Station Clustering using DBSCAN & scikit-learn
DBSCAN is specially very good for tasks like class identification on a spatial context. The wonderful attribute of DBSCAN algorithm is that it can find out any arbitrary shape cluster without getting affected by noise. For example, this following example cluster the location of weather stations in Canada.
DBSCAN can be used here, for instance, to find the group of stations which show the same weather condition. As you can see, it not only finds different arbitrary shaped clusters, can find the denser part of data-centered samples by ignoring less-dense areas or noises.
let's start playing with the data. We will be working according to the following workflow: </font>
About the dataset¶
Environment Canada Monthly Values for July - 2015
| Name in the table | Meaning |
|---|---|
| Stn_Name | Station Name |
| Lat | Latitude (North+, degrees) |
| Long | Longitude (West - , degrees) |
| Prov | Province |
| Tm | Mean Temperature (°C) |
| DwTm | Days without Valid Mean Temperature |
| D | Mean Temperature difference from Normal (1981-2010) (°C) |
| Tx | Highest Monthly Maximum Temperature (°C) |
| DwTx | Days without Valid Maximum Temperature |
| Tn | Lowest Monthly Minimum Temperature (°C) |
| DwTn | Days without Valid Minimum Temperature |
| S | Snowfall (cm) |
| DwS | Days without Valid Snowfall |
| S%N | Percent of Normal (1981-2010) Snowfall |
| P | Total Precipitation (mm) |
| DwP | Days without Valid Precipitation |
| P%N | Percent of Normal (1981-2010) Precipitation |
| S_G | Snow on the ground at the end of the month (cm) |
| Pd | Number of days with Precipitation 1.0 mm or more |
| BS | Bright Sunshine (hours) |
| DwBS | Days without Valid Bright Sunshine |
| BS% | Percent of Normal (1981-2010) Bright Sunshine |
| HDD | Degree Days below 18 °C |
| CDD | Degree Days above 18 °C |
| Stn_No | Climate station identifier (first 3 digits indicate drainage basin, last 4 characters are for sorting alphabetically). |
| NA | Not Available |
1-Download data¶
Did you know? When it comes to Machine Learning, you will likely be working with large datasets. As a business, where can you host your data? IBM is offering a unique opportunity for businesses, with 10 Tb of IBM Cloud Object Storage: Sign up now for free
!wget -O weather-stations20140101-20141231.csv https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/ML0101ENv3/labs/weather-stations20140101-20141231.csv
2- Load the dataset¶
import csv
import pandas as pd
import numpy as np
filename='weather-stations20140101-20141231.csv'
#Read csv
pdf = pd.read_csv(filename)
pdf.head(5)
3-Cleaning¶
pdf = pdf[pd.notnull(pdf["Tm"])]
pdf = pdf.reset_index(drop=True)
pdf.head(5)
4-Visualization¶
Please notice that the size of each data points represents the average of maximum temperature for each station in a year.
from mpl_toolkits.basemap import Basemap
import matplotlib.pyplot as plt
from pylab import rcParams
%matplotlib inline
rcParams['figure.figsize'] = (14,10)
llon=-140
ulon=-50
llat=40
ulat=65
pdf = pdf[(pdf['Long'] > llon) & (pdf['Long'] < ulon) & (pdf['Lat'] > llat) &(pdf['Lat'] < ulat)]
my_map = Basemap(projection='merc',
resolution = 'l', area_thresh = 1000.0,
llcrnrlon=llon, llcrnrlat=llat, #min longitude (llcrnrlon) and latitude (llcrnrlat)
urcrnrlon=ulon, urcrnrlat=ulat) #max longitude (urcrnrlon) and latitude (urcrnrlat)
my_map.drawcoastlines()
my_map.drawcountries()
# my_map.drawmapboundary()
my_map.fillcontinents(color = 'white', alpha = 0.3)
my_map.shadedrelief()
# To collect data based on stations
xs,ys = my_map(np.asarray(pdf.Long), np.asarray(pdf.Lat))
pdf['xm']= xs.tolist()
pdf['ym'] =ys.tolist()
#Visualization1
for index,row in pdf.iterrows():
# x,y = my_map(row.Long, row.Lat)
my_map.plot(row.xm, row.ym,markerfacecolor =([1,0,0]), marker='o', markersize= 5, alpha = 0.75)
#plt.text(x,y,stn)
plt.show()