Installing IBM Spectrum LSF Predictor

Install LSF Predictor.

Before you begin

IBM Spectrum LSF Predictor uses AutoAI in IBM Cloud Pak for Data 4.0 as the training engine. It is required to have Cloud Pak for Data installed and running for LSF Predictor. For more information, see Preparing IBM Cloud Pak for Data.

Consider the following prerequisites when selecting a host for installing LSF Predictor:

LSF Predictor supports the following Linux® operating systems running on x64 platforms:
- Red Hat Enterprise Linux 7.2, or later.
- CentOS 7.2, or later.
- Ubuntu 18.04.
Docker CE (docker-ce) 19.03, or later, must be installed and running.
The following ports must not be in use: 2222, 5050, 9200, 20022.
Internet access to download the Docker-in-Docker (dind) and CentOS 7 (centos:7) images for starting the LSF prediction Docker container.
LSF server or client in a production environment to directly load the LSF configuration files.
The intended host for LSF Predictor must have at least 32 GB of memory.
The installation directory that you use to extract the installation package will be the working directory. Therefore, ensure that the installation directory has at least 100 GB of free space.
Note: The installation directory must be specified as an absolute path and cannot include a symbolic link in the path.
The installer assumes that you agreed to the license terms before starting the LSF Predictor service. View the license agreement in the License/ directory within the installation package.

Procedure

Log in to the intended Linux host as root.
Check that the maximum virtual memory setting is larger than 262144 by running the following command:
```
# sysctl -n vm.max_map_count
```
Add the user for LSF Predictor to the docker user group.
For example, to add lsfadmin to the docker user group,
```
# usermod -aG docker lsfadmin
```
Log back in to the intended host as the user for LSF Predictor.

For example, log in to the intended host as lsfadmin.
Navigate to the intended installation directory and extract the lsf-cognitive-predictor_v1 installation package.
For example,
```
$ cd /opt/ibm
```
```
$ tar -zxvf lsf_cognitive_predictor_v1.tar.gz
```
To start the service as lsfadmin or root, change to the installation directory by running the following command:
```
$ cd lsf_cognitive_v1/
```
Start the LSF Predictor service.
For example, start the LSF Predictor service with directory mapping like the following:
```
$ ./bcogn start -v "LSF_TOP_OUTSIDE_CONTAINER:LSF_TOP_INSIDE_CONTAINER"
```
where LSF_TOP_OUTSIDE_CONTAINER is the real LSF cluster top level file path and LSF_TOP_INSIDE_CONTAINER is the mapping point from inside the Docker container that is created automatically when starting the prediction. You must specify the same file path both times.
For more information about the bcogn start command, refer to Managing the IBM Spectrum LSF Predictor service.
Log in to the LSF Predictor Console from a web browser.
1. Import the LSF Predictor certificate to the web browser as a Trusted Root Certificate.
  
  The certificate is located in lsf_cognitive_v1/config/https/cacert_lsf.pem.
2. From the web browser, log in to the LSF Predictor Console.
  
  https://<lsf_predictor_host>:5050
  
  Enter Admin as both the default user name and password.
  Important: For security reasons, it is highly recommended that you change the Admin password after the first web console login.
  
  Figure 1. Changing Admin password after first web console login

Check that the t_shirt_size parameter value is set to l for "large" in the lsf_cognitive_v1/config/config.json file.


{
"debug": false,
"feature": {
"predictor": {
"enabled": false,
"url": "",
"username": "",
"password": "",
"project": "",
"version": "3.5",
"sample_rate": 1,
"max_samples": 2000000,
"RegressionAlgorithms":["RandomForestRegressorEstimator", "ExtraTreesRegressorEstimator", "XGBRegressorEstimator", "GradientBoostingRegressorEstimator"],
"ClassificationAlgorithms":["ExtraTreesClassifierEstimator", "GradientBoostingClassifierEstimator", "DecisionTreeClassifierEstimator", "RandomForestClassifierEstimator"],
"ranking_metric_regression": "MAE",
"ranking_metric_classification":"Accuracy",
"transformers": ["sum","max"],
"test_size": 0.1,
"t_shirt_size": "l"
},
"optimizer": {
"enabled": false
}
}
}