Metadata-Version: 2.1
Name: dowml
Version: 1.0.0
Summary: Submit existing Decision Optimization instances to WML
Home-page: https://github.com/IBMDecisionOptimization/dowml
Author: Xavier Nodet
Author-email: xavier.nodet@fr.ibm.com
License: UNKNOWN
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3.8
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Operating System :: OS Independent
Classifier: Development Status :: 4 - Beta
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Software Development :: Libraries
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE

# dowml
A library and command line client to use Decision Optimization on WML

## tldr;

```
$ pip install dowml
$ cat my_credentials.txt
{
    'apikey': '<apikey>',
    'url': 'https://us-south.ml.cloud.ibm.com',
    'cos_resource_crn' = 'crn:v1:bluemix:public:cloud-object-storage:global:a/76260f9...',
    'ml_instance_crn': 'crn:v1:bluemix:public:pm-20:eu-de:a/76260f...'
}
$ dowml -w my-credentials.txt
dowml> solve examples/afiro.mps
dowml> wait
dowml> log
dowml> exit
```

## Introduction

The class `DOWMLLib` provides an API to upload Decision Optimization models (CPLEX, CP Optimizer, OPL or docplex) to WML, check their status, and download results.  The script `dowml.py` is an interactive program on top of that library.

In order to use either of them, you need to provide IBM Cloud credentials.
1. By default, `DOWMLLib` (and therefore the Interactive) look for these credentials in an environment variable named `DOWML_CREDENTIALS`. This variable shoud have a value looking like
   ```
   {
       'apikey': '<apikey>',
       'url': 'https://us-south.ml.cloud.ibm.com',
       'cos_resource_crn' = 'crn:v1:bluemix:public:cloud-object-storage:global:a/76260f9...',
       'ml_instance_crn': 'crn:v1:bluemix:public:pm-20:eu-de:a/76260f...',
   }
   ```
   See below for how/where to get these credentials.
2. As an alternative, you can specify a file name as argument to `DOWMLLib.__init__`. The credentials will then be read from that file instead of the environment variable. Accordingly, the Interactive has a command line option `-w` (or `--wml-cred-file`) that must be followed by the path of the file.

Here's a sample session:
```
$ dowml -h
usage: interactive.py [-h] [--wml-cred-file WML_CRED_FILE] [--verbose]
                      [--commands [COMMANDS [COMMANDS ...]]] [--input] [--space SPACE]

Interactive program for DO on WML

optional arguments:
  -h, --help            show this help message and exit
  --wml-cred-file WML_CRED_FILE, -w WML_CRED_FILE
                        Name of the file from which to read WML credentials. If not specified,
                        credentials are read from environment variable $DOWML_CREDENTIALS.
  --verbose, -v         Verbose mode. Causes the program to print debugging messages about its
                        progress. Multiple -v options increase the verbosity. The maximum is 4.
  --commands [COMMANDS [COMMANDS ...]], -c [COMMANDS [COMMANDS ...]]
                        Carries out the specified commands. Each command is executed as if it had
                        been specified at the prompt. The program stops after last command, unless
                        --input is used.
  --input, -i           Prompts for new input commands even if some commands have been specified
                        as arguments using --commands.
  --space SPACE, -s SPACE
                        Id of the space to connect to. Takes precedence over the one specified in
                        the credentials under the 'space_id' key, if any.
$
$
$ dowml -c help type size 'inline yes' 'solve examples/afiro.mps' jobs wait jobs log 'type docplex' 'solve examples/markshare.py examples/markshare1.mps.gz' wait jobs output 'shell ls -l *-*-*-*-*'

Decision Optimization in WML Interactive, version 0.9.0.
Submit and manage Decision Optimization models interactively.
(c) Copyright IBM Corp. 2021

Type ? for a list of commands.

Most commands need an argument that can be either a job id, or the number
of the job, as displayed by the 'jobs' command.  If a command requires a
job id, but none is specified, the last one is used.

dowml> help

Documented commands (type help <topic>):
========================================
cancel  details  help    jobs  output  size   time  version
delete  exit     inline  log   shell   solve  type  wait

dowml> type
Current model type: cplex. Known types: cplex, cpo, opl, docplex
dowml> size
Current size: S. Known sizes: S, M, XL
dowml> inline yes
dowml> solve examples/afiro.mps
Job id: 60c885c9-72ae-4568-be32-1e7c702252c0
dowml> jobs
     #  status      id                                    creation date        type     ver.   size  inputs
=>   1: queued      60c885c9-72ae-4568-be32-1e7c702252c0  2021-08-11 15:07:03  cplex    20.1   S     afiro.mps
dowml> wait
dowml> jobs
     #  status      id                                    creation date        type     ver.   size  inputs
=>   1: completed   60c885c9-72ae-4568-be32-1e7c702252c0  2021-08-11 15:07:03  cplex    20.1   S     afiro.mps
dowml> log
[2021-08-11T13:07:33Z, INFO] CPLEX version 20010000
[2021-08-11T13:07:34Z, WARNING] Changed parameter CPX_PARAM_THREADS from 0 to 1
[2021-08-11T13:07:34Z, INFO] Param[1,067] = 1
[2021-08-11T13:07:34Z, INFO] Param[1,130] = UTF-8
[2021-08-11T13:07:34Z, INFO] Param[1,132] = -1
[2021-08-11T13:07:34Z, INFO]
[2021-08-11T13:07:34Z, INFO] Selected objective sense:  MINIMIZE
[2021-08-11T13:07:34Z, INFO] Selected objective  name:  obj
[2021-08-11T13:07:34Z, INFO] Selected RHS        name:  rhs
[2021-08-11T13:07:34Z, INFO] Version identifier: 20.1.0.0 | 2020-11-10 | 9bedb6d68
[2021-08-11T13:07:34Z, INFO] CPXPARAM_Threads                                 1
[2021-08-11T13:07:34Z, INFO] CPXPARAM_Output_CloneLog                         -1
[2021-08-11T13:07:34Z, INFO] CPXPARAM_Read_APIEncoding                        "UTF-8"
[2021-08-11T13:07:34Z, INFO] Tried aggregator 1 time.
[2021-08-11T13:07:34Z, INFO] LP Presolve eliminated 9 rows and 10 columns.
[2021-08-11T13:07:34Z, INFO] Aggregator did 7 substitutions.
[2021-08-11T13:07:34Z, INFO] Reduced LP has 11 rows, 15 columns, and 37 nonzeros.
[2021-08-11T13:07:34Z, INFO] Presolve time = 0.00 sec. (0.03 ticks)
[2021-08-11T13:07:34Z, INFO]
[2021-08-11T13:07:34Z, INFO] Iteration log . . .
[2021-08-11T13:07:34Z, INFO] Iteration:     1   Scaled dual infeas =             1.200000
[2021-08-11T13:07:34Z, INFO] Iteration:     5   Dual objective     =          -464.753143
[2021-08-11T13:07:34Z, INFO] There are no bound infeasibilities.
[2021-08-11T13:07:34Z, INFO] There are no reduced-cost infeasibilities.
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) Ax-b resid.          = 1.77636e-14 (1.77636e-14)
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) c-B'pi resid.        = 5.55112e-17 (5.55112e-17)
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) |x|                  = 500 (500)
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) |slack|              = 500 (500)
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) |pi|                 = 0.942857 (1.88571)
[2021-08-11T13:07:34Z, INFO] Max. unscaled (scaled) |red-cost|           = 10 (10)
[2021-08-11T13:07:34Z, INFO] Condition number of scaled basis            = 1.5e+01
[2021-08-11T13:07:34Z, INFO] optimal (1)
dowml> type docplex
dowml> solve examples/markshare.py examples/markshare1.mps.gz
Job id: e81b392d-38ed-4d2a-912b-ff0249caf9e7
dowml> wait
[2021-08-11T13:08:09Z, WARNING] Support for Python 3.7 is now enabled and used as the default.
[2021-08-11T13:08:10Z, INFO] Reading markshare1.mps.gz...
dowml> jobs
     #  status      id                                    creation date        type     ver.   size  inputs
     1: completed   60c885c9-72ae-4568-be32-1e7c702252c0  2021-08-11 15:07:03  cplex    20.1   S     afiro.mps
=>   2: completed   e81b392d-38ed-4d2a-912b-ff0249caf9e7  2021-08-11 15:07:44  docplex  20.1   S     markshare.py, markshare1.mps.gz
dowml> output
Storing e81b392d-38ed-4d2a-912b-ff0249caf9e7/solution.json
Storing e81b392d-38ed-4d2a-912b-ff0249caf9e7/kpis.csv
Storing e81b392d-38ed-4d2a-912b-ff0249caf9e7/stats.csv
Storing e81b392d-38ed-4d2a-912b-ff0249caf9e7/log.txt
Storing e81b392d-38ed-4d2a-912b-ff0249caf9e7/details.json
dowml> shell ls -l *-*-*-*-*
-rw-rw-r--  1 nodet  staff  5445 Aug 11 15:08 details.json
-rw-rw-r--  1 nodet  staff    39 Aug 11 15:08 kpis.csv
-rw-rw-r--  1 nodet  staff  7142 Aug 11 15:08 log.txt
-rw-rw-r--  1 nodet  staff  1770 Aug 11 15:08 solution.json
-rw-rw-r--  1 nodet  staff   342 Aug 11 15:08 stats.csv
```


## WML credentials

The DOWML client requires some information in order to connect to the Watson
Machine Learning service.  Two pieces of information are required, and the others
are optional.

### Required items

1. The `apikey` is a secret that identifies the IBM Cloud user. One typically creates
   one key per application or service, in order to be able to revoke them individually
   if needed.
   To generate such a key, open https://cloud.ibm.com/iam/apikeys, and click the blue
   'Create an IBM Cloud API key' on the right.

2. The `url` is the base URL for the REST calls to WML.  The possible values are
   found in https://cloud.ibm.com/apidocs/machine-learning, and depend on which
   region you want to use.

### Optional items

Watson Studio and Watson Machine Learning use _spaces_ to group together, and
isolate from each other, the assets that belong to a single project.  These assets 
include the data files submitted, the results of the jobs, and the _deployments_
(software and hardware configurations) that run these jobs.

The DOWML client will connect to the space specified by the user using
either the `--space` command-line argument or the `space_id` item in the credentials.
If neither of these are specified, the client will look for a space named 
_DOWMLClient-space_, and will try to create such a space if one doesn't exist.
To create a new space, the DOWML client will need both `cos_resource_crn` and
`ml_instance_crn` to have been specified in the credentials.

3. `space_id`: identifier of an existing space to connect to.  Navigate to the 
   'Spaces' tab of your Watson Studio site (e.g. 
   https://eu-de.dataplatform.cloud.ibm.com/ml-runtime/spaces if you are using
   the instance in Germany), right-click on the name of an existing space to
   copy the link. The id of the space is the string of numbers, letters and dashes
   between the last `/` and the `?`.

3. `cos_resource_crn`: WML needs to store some data in a Cloud Object Storage 
   instance.  Open
   https://cloud.ibm.com/resources and locate the 'Storage' section.  Create an
   instance of the Cloud Object Storage service if needed. Once it's listed on
   the resource page, click anywhere on the line for that service, except on its
   name.  This will open a pane on the right which lists the CRN.  Click on the
   symbol at the right to copy this information.  This item is required only for 
   the DOWML client to be able to create a space.  If you specified a `space_id`,
   it is not required.
   
4. `ml_instance_crn`: similarly, you need to identify an instance of Machine 
   Learning service to use
   to solve your jobs.  In the same page https://cloud.ibm.com/resources, open the
   'Services' section.  The 'Product' columns tells you the type of service.  If
   you don't have a 'Machine Learning' instance already, create one.  Then click
   on the corresponding line anywhere except on the name, and copy the CRN displayed
   in the pane that open on the right.  This item is required only for 
   the DOWML client to be able to create a space.  If you specified a `space_id`,
   it is not required.

## Using data assets in Watson Studio

The DOWML library has two modes of operation with respect to sending the models
to the WML service: inline data, or using data assets in Watson Studio.  By default,
data assets are used. This can be changed with the `inline` command.

With inline data, the model is sent directly to the WML service in the _solve_
request itself.  This is the simplest, but it has a number of drawbacks:

- Sending a large model may take a long time, because of network throughput.  Sending
a very large REST request is not at all guaranteed to succeed.

- When solving several times the same model (e.g. to evaluate different parameters),
the model has to be sent each time.

- In order to display the names of the files that were sent, the _jobs_ command
needs to request this information, and it comes with the content of the files
  themselves.  In other words, every _jobs_ command requires downloading the content
  of all the files for all the jobs that exist in the space.

Using data assets in Watson Studio as an intermediate step alleviate all these issues:

- Once the model has been uploaded to Watson Studio, it will be reused for
subsequent jobs without the need to upload it again.

- The job requests refer to the files indirectly, via URLs.  Therefore, they don't
take much space, and listing the jobs doesn't imply to download the content of the
  files.

- Uploading to Watson Studio is done through specialized code that doesn't just send a single
request.  Rather, it divides the upload in multiple reasonably sized chunks that each
  are uploaded individually, with restart if necessary.  Uploading big files is
  therefore much less prone to failure.



