To deploy a Decision
Optimization model, create a model ready for deployment in
your deployment space and then upload your model as an archive. When deployed, you can submit jobs
to your model and monitor job states.
Before you begin
You must have an IBM Cloud account. See https://www.ibm.com/cloud/.
- Log in to IBM
Cloud.
- Create your API
key. Copy or download it from the API key successfully created open
window (you cannot access it again when you close this window).
- Create or select a Machine Learning
service. Copy the service instance name, GUID,
and CRN from the information pane for your instance in the
Resource List>Services view on IBM Cloud. (Expand the list of
services in the Resource List window. Click anywhere in the row next to your
Machine Learning Service name, but not on the name itself. The information pane then opens in the
same window.)
- Create or select a Cloud Object Storage. Copy the Cloud Object Storage
instance name and CRN from the information pane for your instance
in the Resource List>Storage view on IBM Cloud.
- Create a deployment space, from the user interface. Then view it and copy your Space
ID from the settings
tab.
For more information, see Deployment spaces.
About this task
These instructions assume that you have already built your Decision
Optimization model.
Procedure
To deploy a Decision
Optimization
model:
- Package your Decision
Optimization model formulation, with any settings or master
data files, ready for deployment as a
tar.gz
, .zip
, or
.jar
file. Your archive can include the following optional files:
- Your model files
For Python or OPL models, these files usually contain your model formulation.
However, you can also choose to omit these files, especially when your model and data are integrated
in the same file. For example, for CPLEX (.lp
files), CPO (.cpo
files), and mps
format models, to avoid resolving the same mode, you can send these
files later in the job.
- Settings files
For more information, see Solve parameters.
- Master data
These files contain any data that can be reused by all jobs. Including such data
files can make deployment jobs more efficient. For example, you can include a data file for values
that remain constant, such as distances between towns. You can include this data file in the
deployment so that you provide it only once and not at each job request.
Note: For Python models with multiple .py files, put all files in the same
folder in your archive. The same folder must contain a main file called
main.py. Do not use subfolders.
-
Create a model ready for deployment in Watson Machine Learning
providing the following information:
- Machine Learning service
instance
- Deployment space instance
- the hardware specification for the available configuration sizes (small S, medium M,
large L, extra large XL). See configurations.
- Software specification (Decision
Optimization runtime version):
do_22.1
runtime is based on CPLEX 22.1.1.0
do_20.1
runtime is based on CPLEX 20.1.0.1
You can extend the software specification that is provided by Watson Machine Learning. See the ExtendWMLSoftwareSpec notebook in the
jupyter folder of the DO-samples.
Updating CPLEX runtimes:
If you previously deployed your model with a CPLEX runtime that is no longer supported, you can
update your existing deployed model by using either the REST API or the UI.
- The model type:
- opl (do-opl_<runtime version>)
- cplex (do-cplex_<runtime version>)
- cpo (do-cpo_<runtime version>)
- docplex (do-docplex_<runtime version>) using Python 3.11 or 3.10
(deprecated)
(The runtime version can be one of the available Decision
Optimization runtime versions so, for example, an opl model with runtime 22.1 has the model type do-opl_22.1.)
You obtain a MODEL-ID. Your Watson Machine Learning model can then be used in one or multiple
deployments.
- Upload your model archive (
tar.gz
, .zip
, or
.jar
file) on Watson Machine Learning. See Model input and output data file formats for Decision Optimization for information about input file types.
- Deploy your model by using the MODEL-ID, SPACE-ID,
and the hardware specification for the available configuration sizes (small S, medium M,
large L, extra large XL). See configurations. You can also specify the number of nodes to be used (the default
value is 1).
You obtain a DEPLOYMENT-ID.
- Monitor the deployment by using the DEPLOYMENT-ID. Deployment
states can be:
initializing
, updating
, ready
,
or failed
.
Example
See the Deploying a DO model with WML sample for an example of how to deploy
a Decision
Optimization model, create and monitor jobs, and get solutions by using the
Watson Machine Learning Python Client. This notebook uses the diet sample for the Decision
Optimization model and takes you through the whole procedure without using the Decision
Optimization
experiment UI. This sample and the
RunDeployedModel and ExtendWMLSoftwareSpec
notebooks are located in the
jupyter folder of the DO-samples. Select the relevant
product and version subfolder. When downloaded, you
can add these Jupyter notebooks to your project.
See also the REST API example example.