Lab Guide-AI-powered Smart Site Selection V1

Change History
Course Code Applicable To Product Version Issue
Developed By Time Reviewed By New/Update What's New
Zhu Yingying/WX522971 2019.12.11 New

AI-powered Smart Site Selection
Lab Guide
Huawei Technologies Co., Ltd.

AI-powered Smart Site Selection Page 3
1 Lab Environment
1.1 Introduction
1.1.1 About This Test
In this test, HUAWEI CLOUD ModelArts is used for model development and training, and the
model training result is sent to the application service device (development board).
1.1.2 Test Environment

ModelArts is a one-stop development platform, including data processing, model training,
management, and deployment, and AI Market for sharing models, APIs, and datasets.
Figure 1-1 ModelArts architecture
1.1.3 Environment Preparation

Step 1 Log in to the HUAWEI CLOUD website at https://www.huaweicloud.com/en-us/.
Before logging in to HUAWEI CLOUD, register an account at

https://www.huaweicloud.com/en-us/. Click Log In and enter the username and password as
prompted.
Step 2 Log in to management console.
Click Console in the upper right corner of the page. The console page is displayed.
Step 3 Create an Object Storage Service (OBS) bucket.
In the service list on the left, choose Storage > Object Storage Service.
In the upper right corner of the page, click Create Bucket to create a bucket for storing the
data and configuration files required in this test.
Set the bucket name, for example, aiot2. Retain the default values for other parameters.
After the setting is complete, click Create Now.
Return to the Object Storage Service page. If the bucket name created in the previous step
exists in the bucket name list, the bucket is successfully created.
Step 4 Upload files.
In the bucket list, click the name of the created bucket. On the aiot2 page, click the Objects
tab. ModelArts cannot be directly associated with the root directory of OBS. Therefore, you
need to create a data folder to store the files required in this test.
Go to the data folder and click Upload Object.
Select the data files and configuration files to be uploaded.
If the uploaded files exist in the object list, the upload is successful.
Step 5 Create a ModelArts notebook instance.
Choose Service List > EI Enterprise Intelligence > ModelArts.

Choose DevEnviron > Notebooks and click Create.

Step 6 Configure a ModelArts instance.
The instance name can be customized, for example, notebook-AIoT. Select Python3 as the
work environment and OBS as the storage, and set the data storage path.
If the created instance name exists in the notebook instance list, the creation is successful.
Step 7 Create a notebook.
In this test, methods in the sklearn library are used. Therefore, select TensorFlow-1.13.1
during the creation.
Step 8 Synchronize data using the Sync OBS function.
The Sync OBS function is used to synchronize the objects selected in the list of notebook
instance files from the OBS bucket to the current container directory ~/work. Select the data
files to be synchronized and click Sync OBS. In the dialog box that is displayed, select YES.
If information similar to the following is displayed, the data synchronization is successful.

----End
2 Distribution Warehouse Location Selection
2.1 Introduction
2.1.1 About This Test
Offerings need to be supplemented for vending machines. How can we determine the
distribution warehouse location to ensure economical and efficient offering delivery? This
section uses the k-means clustering method to perform clustering on the location (longitude
and latitude) of each device to determine the optimal location of the distribution warehouse.
2.1.2 Objectives
On completion of this test, you will be able to master:
 Important parameters of clustering models
 Principles of the k-means clustering method
 K-means clustering modeling method
2.2 Procedure
2.2.1 Importing Data
Import data file details.csv.
import pandas as pd
data = pd.read_csv('./details.csv')
#Adjust the path based on site requirements.
print(data.shape)#Check whether the number of samples is correct.
2.2.2 Processing Data

Extract the longitude and latitude information about the device from the raw data for
clustering.
addr = data[['longitude','latitude']]
addr.head()
Part of the data:

2.2.3 Clustering Data

Step 1 Convert the data format.
The data input format required by sklearn.cluster is array. If the DataFrame data format is
directly used, an error is reported. Therefore, you need to convert the arr format to the array
format.
addr_1=addr.values
addr_1 #View the data.
Part of the data:

array([[120.82, 29.58],
[120.57, 30.63],
[120.27, 30.17],
...,
[120.28, 28.15],
[120.3 , 30.42],
[120.2 , 30.27]])
Step 2 Create a clustering model.
Invoke k-means in the sklearn.cluster library to implement clustering for device addresses.
The n_cluster parameter indicates the number of clusters and is manually input based on site
requirements. random_state indicates the initial random number sequence specified during
initial cluster center selection.
from sklearn.cluster import KMeans
sites_kmeans = KMeans(n_clusters=3,random_state = 12) #Set three clusters.

sites_kmeans.fit(addr_1)
Parameters of the KMeans() method are output:

KMeans(algorithm='auto', copy_x=True, init='k-means++', max_iter=300,
n_clusters=3, n_init=10, n_jobs=None, precompute_distances='auto',
random_state=123, tol=0.0001, verbose=0)
2.2.4 Evaluating the Clustering Effect

Step 1 Evaluate the clustering effect.
Use the silhouette_score() method in the sklearn.metrics package to evaluate the clustering
effect.
from sklearn.metrics import silhouette_score
silhouette_score(addr_1,sites_kmeans.labels_,sample_size=1000)
The output is as follows:

0.46196462466163146
Step 2 Compare the clustering effects of different k values.
Set the number of clusters to 2, 4, and 5, respectively.

##Two clusters
sites_kmeans_2 = KMeans(n_clusters=2,random_state = 12)
sites_kmeans_2.fit(addr_1)
##Four clusters
##Five clusters
View the clustering effect.

print("when the number of clusters is 2",silhouette_score(addr_1,sites_kmeans_2.labels_, sample_size=1000))

When the number of clusters is 2 0.4488362943453525
The result shows that when the number of clusters is 4, the model evaluation result is the
best. If the construction cost is sufficient, four distribution warehouses are recommended.
----End
2.2.5 Viewing Results

Step 1 Print the k-means output.
The output includes:
cluster_centers_: coordinates of the cluster center
Labels_: classification result label of each point
inertia_: sum of the distance from each point to the centroid of the cluster.
print('Labels:',sites_kmeans_4.labels_[:5])
print('Prec:',sites_kmeans_4.predict(addr_1)[:5])
print('Centers:\n',sites_kmeans_4.cluster_centers_)

Labels: [2 0 0 1 2]
Prec: [2 0 0 1 2]
Centers:
[[120.24059235 30.34716132]
[120.70759845 28.07876567]
[121.64831582 29.86180276]
[119.31063156 28.92008776]]
# Save the longitude and latitude information, which can be used in DLV display.
Step 2 Visualize the results.
Devices are divided into four clusters based on their longitudes and latitudes to display the
clustering result of each device.
plt.figure(figsize=[12,9])
for cluster,marker in zip(range(4),['^','o','+','*']):

x_axis = addr_1[:,0][sites_kmeans_4.labels_ == cluster]
y_axis = addr_1[:,1][sites_kmeans_4.labels_ == cluster]
plt.scatter(x_axis,y_axis,marker = marker)
plt.show()
#^ indicates the drawing triangle, o indicates the drawing circle, and + indicates the drawing +.
#Obtain the first column of addr and assign the cluster to x.labels_.
----End

Lab Guide-AI-powered Smart Site Selection V1

Uploaded by

Copyright:

Available Formats

Lab Guide-AI-powered Smart Site Selection V1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lab Guide-AI-powered Smart Site Selection V1

Uploaded by

Copyright:

Available Formats

Change History

Course Code Applicable To Product Version Issue

Developed By Time Reviewed By New/Update What's New

Zhu Yingying/WX522971 2019.12.11 New

Huawei Technologies Co., Ltd.

1.1.2 Test Environment

Figure 1-1 ModelArts architecture

1.1.3 Environment Preparation

Before logging in to HUAWEI CLOUD, register an account at

Step 2 Log in to management console.

Step 3 Create an Object Storage Service (OBS) bucket.

Step 4 Upload files.

Go to the data folder and click Upload Object.

Select the data files and configuration files to be uploaded.

Step 5 Create a ModelArts notebook instance.

Choose Service List > EI Enterprise Intelligence > ModelArts.

Choose DevEnviron > Notebooks and click Create.

Step 6 Configure a ModelArts instance.

Step 7 Create a notebook.

Step 8 Synchronize data using the Sync OBS function.

If information similar to the following is displayed, the data synchronization is successful.

2 Distribution Warehouse Location Selection

2.2.2 Processing Data

Part of the data:

2.2.3 Clustering Data

Part of the data:

Step 2 Create a clustering model.

sites_kmeans = KMeans(n_clusters=3,random_state = 12) #Set three clusters.

Parameters of the KMeans() method are output:

2.2.4 Evaluating the Clustering Effect

The output is as follows:

Step 2 Compare the clustering effects of different k values.

Set the number of clusters to 2, 4, and 5, respectively.

View the clustering effect.

The output is as follows:

2.2.5 Viewing Results

The output includes:

cluster_centers_: coordinates of the cluster center

Labels_: classification result label of each point

The output is as follows:

Step 2 Visualize the results.

for cluster,marker in zip(range(4),['^','o','+','*']):

The output is as follows:

You might also like