Jump to: navigation, search

Meteos/ExampleLinear

< Meteos
Revision as of 06:45, 10 January 2017 by HiroyukiEguchi (talk | contribs) (→‎6. Predict)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Predict a Sales Figures using Meteos

In this example, you create a prediction model which predict sales by using Linear Regression.

Linear Regression is one of the algorithms in supervised learning.

Usecase.png

1. Create a experiment template

Create template of experiment. Experiment is a workspace of Machine Learning.

You have to confirm a glance image id of meteos image, and a neutron network id before creating a template.

You can use a format located in python-meteosclient/sample/json/template.json 

$ glance image-list | grep meteos
| a6b7de0b-10ff-462c-9c86-25c8a5622a57 | meteos                          |

$ neutron net-list | grep public
| 4222b557-6d9f-405c-b1ff-0f454d2f35bf | public  | 1f979ae3-d6b7-4d03-ba0f-9d9112581783             |

$ vim sample/json/template.json

$ cat sample/json/template.json
{
    "display_name": "example-template",
    "display_description": "This is a sample template of experiment",
    "image_id" : "a6b7de0b-10ff-462c-9c86-25c8a5622a57",
    "master_nodes_num": 1,
    "master_flavor_id": "4",
    "worker_nodes_num": 2,
    "worker_flavor_id": "2",
    "spark_version": "1.6.0",
    "floating_ip_pool": "4222b557-6d9f-405c-b1ff-0f454d2f35bf"
}

$ meteos template-create --json sample/json/template.json
+---------------+-----------------------------------------+
| Property      | Value                                   |
+---------------+-----------------------------------------+
| cluster_id    | None                                    |
| created_at    | 2016-12-15T22:55:03.000000              |
| description   | This is a sample template of experiment |
| id            | d3180a78-74cc-432d-9e9d-68640b18abae    |
| master_flavor | 4                                       |
| master_nodes  | 1                                       |
| name          | example-template                        |
| project_id    | 4034bb3cd9324776a68c140fdd18baa4        |
| spark_version | 1.6.0                                   |
| status        | available                               |
| user_id       | 64ad04e101df40b4b51e5f370a83412c        |
| worker_flavor | 2                                       |
| worker_nodes  | 2                                       |
+---------------+-----------------------------------------+

2. Create a experiment from template

Create a experiment by using template created in the above step. You have to confirm a neutron private network id and create keypair before creating a template.

You can use a format located in python-meteosclient/sample/json/experiment.json 

$ nova keypair-add key1 > ~/key1.pem && chmod 600 ~/key1.pem

$ neutron net-list | grep private
| e4fcc49c-48e5-48f8-9599-bb5eba1339c9 | private | e15c24a5-dfdd-4428-b27d-9827b35600c0 10.0.0.0/26 |

$ vim sample/json/experiment.json

$ cat sample/json/experiment.json
{
    "display_name": "example-experiment",
    "display_description": "This is a sample experiment",
    "key_name": "key1",
    "neutron_management_network": "e4fcc49c-48e5-48f8-9599-bb5eba1339c9",
    "template_id": "d3180a78-74cc-432d-9e9d-68640b18abae"
}

$ meteos experiment-create --json sample/json/experiment.json
+--------------------+--------------------------------------+
| Property           | Value                                |
+--------------------+--------------------------------------+
| created_at         | 2016-12-15T22:56:51.000000           |
| description        | This is a sample experiment          |
| id                 | fcc8d055-e801-4652-af8c-5aabedbf0286 |
| key_name           | key1                                 |
| management_network | e4fcc49c-48e5-48f8-9599-bb5eba1339c9 |
| name               | example-experiment                   |
| project_id         | 4034bb3cd9324776a68c140fdd18baa4     |
| status             | creating                             |
| user_id            | 64ad04e101df40b4b51e5f370a83412c     |
+--------------------+--------------------------------------+

Meteos creates a experiment using OpenStack Sahara spark plugin.

You can see a sahara cluster and nova VMs created by Meteos as below. 

$ openstack dataprocessing cluster list (or sahara cluster-list)
+------------------+--------------------------------------+-------------+----------------+----------+
| Name             | Id                                   | Plugin name | Plugin version | Status   |
+------------------+--------------------------------------+-------------+----------------+----------+
| cluster-fcc8d055 | 5736d157-ac7c-41de-8aca-78f7afa7e99c | spark       | 1.6.0          | Spawning |
+------------------+--------------------------------------+-------------+----------------+----------+

$ openstack server list (or nova list)
+--------------------------------------+----------------------------+--------+------------+-------------+------------------+
| ID                                   | Name                       | Status | Task State | Power State | Networks         |
+--------------------------------------+----------------------------+--------+------------+-------------+------------------+
| 48a9f429-7756-4bed-8dd6-6dc6140ef897 | cluster-fcc8d055-master-0  | ACTIVE | -          | Running     | private=10.0.0.5 |
| 88ff2070-dfe8-45da-aa5c-02ac3e9de3b8 | cluster-fcc8d055-workers-0 | ACTIVE | -          | Running     | private=10.0.0.7 |
| a57dfa5d-8b55-47c7-aae7-d5b3c8779787 | cluster-fcc8d055-workers-1 | ACTIVE | -          | Running     | private=10.0.0.4 |
+--------------------------------------+----------------------------+--------+------------+-------------+------------------+

3. Upload a raw data

Upload a raw data (in this example past sales figures data) to OpenStack Swift.

You can use a sample data located in python-meteosclient/sample/data/linear_data.txt

Raw data shows "sales figures", "day", "month", "year", "day of week", "parameter which indicates weather", "degree", "humidity" from left. 

$ cd sample/data/

/sample/data$ head linear_data.txt
4500,1,1,2016,5,0,40,50
8000,2,1,2016,6,1,60,80
9500,3,1,2016,0,2,88,92
5000,4,1,2016,1,3,90,90
0,5,1,2016,2,2,90,80
4500,6,1,2016,3,3,80,90
4000,7,1,2016,4,1,60,80
4500,8,1,2016,5,0,40,50
8000,9,1,2016,6,0,30,50
9500,10,1,2016,0,0,40,50

/sample/data$ swift upload meteos linear_data.txt
linear_data.txt

4. Parse a raw data

Parse a raw data to eliminate exception data.

In this case, you have to eliminate holiday sales figures because exception data affect the accuracy of prediction model.

You can use a sample format located in python-meteosclient/sample/json/dataset_parse.json

You can see the head data of parsed dataset by executing "meteos dataset-show <dataset-uuid>" command. 

$ vim ../python-meteosclient/sample/json/dataset_parse.json

$ cat sample/json/dataset_parse.json
{
    "source_dataset_url": "swift://meteos/linear_data.txt",
    "display_name": "sample-data",
    "display_description": "This is a sample dataset",
    "method": "parse",
    "params": [{"method": "filter", "args": "lambda l: l.split(',')[0] != '0'"}],
    "experiment_id": "fcc8d055-e801-4652-af8c-5aabedbf0286",
    "swift_tenant": "demo",
    "swift_username": "demo",
    "swift_password": "nova"
}

$ meteos dataset-create --json sample/json/dataset_parse.json
+-------------+--------------------------------------+
| Property    | Value                                |
+-------------+--------------------------------------+
| created_at  | 2016-12-15T23:04:11.000000           |
| description | This is a sample dataset             |
| head        | None                                 |
| id          | 91d98f6d-a065-431b-b8e0-1b996ac85cec |
| name        | sample-data                          |
| project_id  | 4034bb3cd9324776a68c140fdd18baa4     |
| status      | creating                             |
| stderr      | None                                 |
| user_id     | 64ad04e101df40b4b51e5f370a83412c     |
+-------------+--------------------------------------+

$ meteos dataset-list
+--------------------------------------+-------------+-----------+--------------------------------+
| id                                   | name        | status    | source_dataset_url             |
+--------------------------------------+-------------+-----------+--------------------------------+
| 91d98f6d-a065-431b-b8e0-1b996ac85cec | sample-data | available | swift://meteos/linear_data.txt |
+--------------------------------------+-------------+-----------+--------------------------------+

You can see that holiday sales figures has been eliminated. 

$ meteos dataset-show 91d98f6d-a065-431b-b8e0-1b996ac85cec
+-------------+--------------------------------------+
| Property    | Value                                |
+-------------+--------------------------------------+
| created_at  | 2016-12-15T23:04:11.000000           |
| description | This is a sample dataset             |
| head        | [u'4500,1,1,2016,5,0,40,50',         |
|             | u'8000,2,1,2016,6,1,60,80',          |
|             | u'9500,3,1,2016,0,2,88,92',          |
|             | u'5000,4,1,2016,1,3,90,90',          |
|             | u'4500,6,1,2016,3,3,80,90',          |
|             | u'4000,7,1,2016,4,1,60,80',          |
|             | u'4500,8,1,2016,5,0,40,50',          |
|             | u'8000,9,1,2016,6,0,30,50',          |
|             | u'9500,10,1,2016,0,0,40,50',         |
|             | u'5000,11,1,2016,1,1,60,80']         |
| id          | 91d98f6d-a065-431b-b8e0-1b996ac85cec |
| name        | sample-data                          |
| project_id  | 4034bb3cd9324776a68c140fdd18baa4     |
| status      | available                            |
| stderr      |                                      |
| user_id     | 64ad04e101df40b4b51e5f370a83412c     |
+-------------+--------------------------------------+

5. Create a prediction model

In this example, User creates a Linear Regression Model from parsed dataset.

Parsed dataset has been already distributed in hdfs of experiment environment.

So, you spefity the internal url (internal://<dataset-id>) in source_dataset_url parameter. 

$ vim sample/json/model_linear.json

$ cat sample/json/model_linear.json
{
    "display_name": "linear-model",
    "display_description": "Sample LinearRegression Model",
    "source_dataset_url": "internal://91d98f6d-a065-431b-b8e0-1b996ac85cec",
    "model_type": "LinearRegression",
    "model_params": "{'numIterations': 200}",
    "experiment_id": "fcc8d055-e801-4652-af8c-5aabedbf0286"
}

$ meteos model-create --json sample/json/model_linear.json
+-------------+--------------------------------------+
| Property    | Value                                |
+-------------+--------------------------------------+
| created_at  | 2016-12-15T23:09:36.000000           |
| description | Sample LinearRegression Model        |
| id          | 3cf02c2f-f043-49e5-a0df-0dc782868312 |
| name        | linear-model                         |
| params      | eydudW1JdGVyYXRpb25zJzogMjAwfQ==     |
| project_id  | 4034bb3cd9324776a68c140fdd18baa4     |
| status      | creating                             |
| stderr      | None                                 |
| stdout      | None                                 |
| type        | LinearRegression                     |
| user_id     | 64ad04e101df40b4b51e5f370a83412c     |
+-------------+--------------------------------------+

$ meteos model-list
+--------------------------------------+--------------+-----------+------------------+-------------------------------------------------+
| id                                   | name         | status    | type             | source_dataset_url                              |
+--------------------------------------+--------------+-----------+------------------+-------------------------------------------------+
| 3cf02c2f-f043-49e5-a0df-0dc782868312 | linear-model | available | LinearRegression | internal://91d98f6d-a065-431b-b8e0-1b996ac85cec |
+--------------------------------------+--------------+-----------+------------------+-------------------------------------------------+

6. Predict

Create a learning job predicting a Sales Figures.

Specify the input value as "args" parameter.

Retrieve a predicted data as a stdout of job execution. 

$ vim sample/json/learning.json

$ cat sample/json/learning.json
{
    "display_name": "example-learning-job",
    "display_description": "This is a sample job",
    "model_id": "3cf02c2f-f043-49e5-a0df-0dc782868312",
    "method": "predict",
    "args": "1,7,2016,5,0,60,80"
}

$ meteos learning-create --json sample/json/learning.json
+-------------+--------------------------------------+
| Property    | Value                                |
+-------------+--------------------------------------+
| args        | MSw3LDIwMTYsNSwwLDYwLDgw             |
| created_at  | 2016-12-15T23:13:29.000000           |
| description | This is a sample job                 |
| id          | ffcd3ae6-8b53-437e-94e2-3aabde741cb0 |
| method      | predict                              |
| name        | example-learning-job                 |
| project_id  | 4034bb3cd9324776a68c140fdd18baa4     |
| status      | creating                             |
| stderr      | None                                 |
| stdout      | None                                 |
| user_id     | 64ad04e101df40b4b51e5f370a83412c     |
+-------------+--------------------------------------+

$ meteos learning-list
+--------------------------------------+----------------------+-----------+--------------------+---------------+
| id                                   | name                 | status    | args               | stdout        |
+--------------------------------------+----------------------+-----------+--------------------+---------------+
| ffcd3ae6-8b53-437e-94e2-3aabde741cb0 | example-learning-job | available | 1,7,2016,5,0,60,80 | 4048.18352502 |
+--------------------------------------+----------------------+-----------+--------------------+---------------+

7. Online Prediction

You can load a Prediction Model in advance for online prediction by using "meteos-load" command.

In online prediction, user can retrieve a predicted data immediately.

Note: Meteos can not load multiple models at the same time. So, you have to unload a model before loading another model. 

$ meteos model-load 3cf02c2f-f043-49e5-a0df-0dc782868312

$ meteos model-list
+--------------------------------------+--------------+------------+------------------+-------------------------------------------------+
| id                                   | name         | status     | type             | source_dataset_url                              |
+--------------------------------------+--------------+------------+------------------+-------------------------------------------------+
| 3cf02c2f-f043-49e5-a0df-0dc782868312 | linear-model | activating | LinearRegression | internal://91d98f6d-a065-431b-b8e0-1b996ac85cec |
+--------------------------------------+--------------+------------+------------------+-------------------------------------------------+

Status is set to "active" when loading is complete. 

$ meteos model-list
+--------------------------------------+--------------+--------+------------------+-------------------------------------------------+
| id                                   | name         | status | type             | source_dataset_url                              |
+--------------------------------------+--------------+--------+------------------+-------------------------------------------------+
| 3cf02c2f-f043-49e5-a0df-0dc782868312 | linear-model | active | LinearRegression | internal://91d98f6d-a065-431b-b8e0-1b996ac85cec |
+--------------------------------------+--------------+--------+------------------+-------------------------------------------------+

You can get a predicted data as a response of REST API . 

$ meteos learning-create --json sample/json/learning.json
+-------------+--------------------------------------+
| Property    | Value                                |
+-------------+--------------------------------------+
| args        | MSw3LDIwMTYsNSwwLDYwLDgw             |
| created_at  | 2016-12-15T23:13:29.000000           |
| description | This is a sample job                 |
| id          | ffcd3ae6-8b53-437e-94e2-3aabde741cb0 |
| method      | predict                              |
| name        | example-learning-job                 |
| project_id  | 4034bb3cd9324776a68c140fdd18baa4     |
| status      | available                            |
| stderr      | None                                 |
| stdout      | 4048.18352502                        |
| user_id     | 64ad04e101df40b4b51e5f370a83412c     |
+-------------+--------------------------------------+

$ meteos model-unload 3cf02c2f-f043-49e5-a0df-0dc782868312
Retrieved from "https://wiki.openstack.org/w/index.php?title=Meteos/ExampleLinear&oldid=149699"