Use Keras and HPO to recognize hand-written digits.ipynb

Kernel: watsonx-ai-samples-py-312

Use Keras and hyperparameter optimization (HPO) to recognize hand-written digits with ibm-watsonx-ai

This notebook contains steps and code to demonstrate support of Deep Learning experiments in the watsonx.ai service. It introduces commands for data retrieval, training definition persistence, experiment training, model persistence, model deployment and scoring.

Some familiarity with Python is helpful. This notebook uses Python 3.12.

Note: To use Deep Learning experiments, you must install and configure the IBM® Software Hub scheduling service.

Learning goals

The learning goals of this notebook are:

• Working with the watsonx.ai service.

• Training Deep Learning models (TensorFlow).

• Saving trained models in watsonx.ai repository.

• Online deployment and scoring of the trained model.

Contents

This notebook contains the following parts:

1. Setup

2. Create model definition

3. Train model

4. Persist trained model

5. Deploy and Score

6. Clean up

7. Summary and next steps

1. Set up the environment

Before you use the sample code in this notebook, you must perform the following setup tasks:

• Contact with your Cloud Pak for Data administrator and ask them for your account credentials

Install dependencies

Note: ibm-watsonx-ai documentation can be found here.

%pip install -U wget | tail -n 1
%pip install -U matplotlib | tail -n 1
%pip install -U ibm-watsonx-ai | tail -n 1

Successfully installed wget-3.2
Successfully installed contourpy-1.3.2 cycler-0.12.1 fonttools-4.59.0 kiwisolver-1.4.8 matplotlib-3.10.3 numpy-2.3.1 pillow-11.3.0 pyparsing-3.2.3
Successfully installed anyio-4.9.0 cachetools-6.1.0 certifi-2025.7.14 charset_normalizer-3.4.2 h11-0.16.0 httpcore-1.0.9 httpx-0.28.1 ibm-cos-sdk-2.14.2 ibm-cos-sdk-core-2.14.2 ibm-cos-sdk-s3transfer-2.14.2 ibm-watsonx-ai-1.3.31 idna-3.10 jmespath-1.0.1 lomond-0.3.3 pandas-2.2.3 pytz-2025.2 requests-2.32.4 sniffio-1.3.1 tabulate-0.9.0 typing_extensions-4.14.1 tzdata-2025.2 urllib3-2.5.0

Define credentials

Authenticate the watsonx.ai Runtime service on IBM Cloud Pak for Data. You need to provide the admin's username and the platform url.

username = "PASTE YOUR USERNAME HERE"
url = "PASTE THE PLATFORM URL HERE"

Use the admin's api_key to authenticate watsonx.ai Runtime services:

import getpass
from ibm_watsonx_ai import Credentials

credentials = Credentials(
    username=username,
    api_key=getpass.getpass("Enter your watsonx.ai API key and hit enter: "),
    url=url,
    instance_id="openshift",
    version="5.2",
)

Alternatively you can use the admin's password:

import getpass
from ibm_watsonx_ai import Credentials

if "credentials" not in locals() or not credentials.api_key:
    credentials = Credentials(
        username=username,
        password=getpass.getpass("Enter your watsonx.ai password and hit enter: "),
        url=url,
        instance_id="openshift",
        version="5.2",
    )

Create APIClient instance

from ibm_watsonx_ai import APIClient

client = APIClient(credentials)

Working with spaces

First of all, you need to create a space that will be used for your work. If you do not have space already created, you can use {PLATFORM_URL}/ml-runtime/spaces?context=icp4data to create one.

• Click New Deployment Space

• Create an empty space

• Go to space Settings tab

• Copy space_id and paste it below

Tip: You can also use SDK to prepare the space for your work. More information can be found here.

Action: Assign space ID below

space_id = "PASTE YOUR SPACE ID HERE"

You can use list method to print all existing spaces.

client.spaces.list(limit=10)

To be able to interact with all resources available in watsonx.ai, you need to set space which you will be using.

client.set.default_space(space_id)

'SUCCESS'

2. Create model definition

For the purpose of this example two Keras model definitions have been prepared:

• Multilayer Perceptron (MLP)

• Convolution Neural Network (CNN)

2.1 Prepare model definition metadata

meta_props = {
    client.model_definitions.ConfigurationMetaNames.NAME: "MNIST mlp model definition",
    client.model_definitions.ConfigurationMetaNames.DESCRIPTION: "MNIST mlp model definition",
    client.model_definitions.ConfigurationMetaNames.COMMAND: "mnist_mlp.py",
    client.model_definitions.ConfigurationMetaNames.PLATFORM: {
        "name": "python",
        "versions": ["3.11"],
    },
    client.model_definitions.ConfigurationMetaNames.VERSION: "2.0",
}

2.2 Get sample model definition content files from git (Python scripts with CNN and MLP)

import os, wget

filename_mnist = "MNIST.zip"

if not os.path.isfile(filename_mnist):
    filename_mnist = wget.download(
        "https://github.com/IBM/watsonx-ai-samples/raw/master/cpd5.2/definitions/keras/mnist/MNIST.zip"
    )

Tip: Convert below cell to code and run it to see model deinition's code.

!unzip -oqd . MNIST.zip && cat mnist_mlp.py

2.3 Publish model definition

model_definition_details = client.model_definitions.store(
    filename_mnist, meta_props=meta_props
)

model_definition_id = client.model_definitions.get_id(model_definition_details)
print(model_definition_id)

ea26fd0e-9cbb-4028-9956-7fd89ad0e176

List model definitions

client.model_definitions.list()

3. Train model

3.1 Prepare training metadata

training_metadata = {
    client.training.ConfigurationMetaNames.NAME: "Keras-MNIST",
    client.training.ConfigurationMetaNames.DESCRIPTION: "Keras-MNIST predict written digits",
    client.training.ConfigurationMetaNames.TRAINING_RESULTS_REFERENCE: {
        "name": "MNIST results",
        "connection": {},
        "location": {"path": f"spaces/{space_id}/assets/experiment"},
        "type": "fs",
    },
    client.training.ConfigurationMetaNames.MODEL_DEFINITION: {
        "id": model_definition_id,
        "hardware_spec": {
            "name": "CPUx4",
            "nodes": 1,
        },  # To use GPU for training, set hardware_spec.name to `GPUx1` for example
        "software_spec": {"name": "tensorflow_rt24.1-py3.11"},
    },
    client.training.ConfigurationMetaNames.TRAINING_DATA_REFERENCES: [
        {
            "name": "training_input_data",
            "type": "fs",
            "connection": {},
            "location": {"path": "tf-mnist"},
            "schema": {
                "id": "idmlp_schema",
                "fields": [{"name": "text", "type": "string"}],
            },
        }
    ],
}

3.2 Train the model

training = client.training.run(training_metadata, asynchronous=False)

##############################################

Running 'c6217edf-76f0-4252-8ff3-0282547fb316'

##############################################


pending....
running................
completed
Training of 'c6217edf-76f0-4252-8ff3-0282547fb316' finished successfully.

3.3 Get training ID

training_id = client.training.get_id(training)

3.4 Get training details

import json

training_details = client.training.get_details(training_id)
print(json.dumps(training_details, indent=2))

{
  "metadata": {
    "created_at": "2025-07-17T07:06:07.278Z",
    "description": "Keras-MNIST predict written digits",
    "id": "c6217edf-76f0-4252-8ff3-0282547fb316",
    "modified_at": "2025-07-17T07:08:18.285Z",
    "name": "Keras-MNIST",
    "space_id": "3b92fb5d-b85f-46ee-8dc6-1660c411d6f4"
  },
  "entity": {
    "model_definition": {
      "hardware_spec": {
        "name": "CPUx4"
      },
      "id": "ea26fd0e-9cbb-4028-9956-7fd89ad0e176",
      "software_spec": {
        "name": "tensorflow_rt24.1-py3.11"
      }
    },
    "results_reference": {
      "connection": {},
      "location": {
        "path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment",
        "model": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/data/model",
        "notebooks_path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/notebooks",
        "training": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316",
        "training_status": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/training-status.json",
        "model_request_path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/assets/c6217edf-76f0-4252-8ff3-0282547fb316/resources/wml_model/request.json",
        "logs": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/logs",
        "assets_path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/assets"
      },
      "type": "fs"
    },
    "space_id": "3b92fb5d-b85f-46ee-8dc6-1660c411d6f4",
    "status": {
      "completed_at": "2025-07-17T07:08:17.989Z",
      "running_at": "2025-07-17T07:06:32.766Z",
      "state": "completed"
    },
    "training_data_references": [
      {
        "location": {
          "path": "tf-mnist"
        },
        "schema": {
          "fields": [
            {
              "name": "text",
              "nullable": false,
              "type": "string"
            }
          ],
          "id": "idmlp_schema"
        },
        "type": "fs"
      }
    ]
  }
}

List trainings

client.training.list()

Cancel training

You can cancel the training run by calling client.training.cancel(training_id).

Tip: If you want to delete train runs and results, add hard_delete=True as a parameter.

4. Persist trained model

4.1 Publish model

software_spec_id = client.software_specifications.get_id_by_name(
    "tensorflow_rt24.1-py3.11"
)

model_meta_props = {
    client.repository.ModelMetaNames.NAME: "Keras MNIST",
    client.repository.ModelMetaNames.TYPE: "tensorflow_2.14",
    client.repository.ModelMetaNames.SOFTWARE_SPEC_ID: software_spec_id,
}

published_model = client.repository.store_model(
    training_id, meta_props=model_meta_props
)
model_id = client.repository.get_model_id(published_model)

4.2 Get model details

model_details = client.repository.get_details(model_id)
print(json.dumps(model_details, indent=2))

{
  "metadata": {
    "name": "Keras MNIST",
    "space_id": "3b92fb5d-b85f-46ee-8dc6-1660c411d6f4",
    "resource_key": "5547a1e0-15da-40cd-8e11-828a1891c444",
    "id": "5d915c60-328b-4773-a8b6-036395644953",
    "created_at": "2025-07-17T07:09:32Z",
    "rov": {
      "member_roles": {
        "1000331001": {
          "user_iam_id": "1000331001",
          "roles": [
            "OWNER"
          ]
        }
      }
    },
    "owner": "1000331001"
  },
  "entity": {
    "model_definition": {
      "id": "ea26fd0e-9cbb-4028-9956-7fd89ad0e176"
    },
    "software_spec": {
      "id": "2c33167d-b11c-5490-a305-3e5e95db5c4d"
    },
    "type": "tensorflow_rt24.1",
    "training_id": "c6217edf-76f0-4252-8ff3-0282547fb316",
    "content_location": {
      "type": "fs",
      "location": {
        "assets_path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/assets",
        "logs": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/logs",
        "model": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/data/model",
        "notebooks_path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/notebooks",
        "path": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment",
        "training": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316",
        "training_status": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/training-status.json"
      },
      "contents": [
        {
          "content_format": "native",
          "location": "spaces/3b92fb5d-b85f-46ee-8dc6-1660c411d6f4/assets/experiment/c6217edf-76f0-4252-8ff3-0282547fb316/data/model/c6217edf-76f0-4252-8ff3-0282547fb316.zip",
          "file_name": "c6217edf-76f0-4252-8ff3-0282547fb316.zip"
        }
      ]
    },
    "content_import_state": "completed"
  }
}

List stored models

client.repository.list_models()

5. Deploy and score

5.1 Create online deployment for published model

deployment = client.deployments.create(
    model_id,
    meta_props={
        client.deployments.ConfigurationMetaNames.NAME: "Keras MNIST",
        client.deployments.ConfigurationMetaNames.ONLINE: {},
    },
)

deployment_id = client.deployments.get_id(deployment)

######################################################################################

Synchronous deployment creation for id: '5d915c60-328b-4773-a8b6-036395644953' started

######################################################################################


initializing
Note: online_url is deprecated and will be removed in a future release. Use serving_urls instead.
........
ready


-----------------------------------------------------------------------------------------------
Successfully finished deployment creation, deployment_id='52a3e6ef-885d-405e-a641-53ca625ea29d'
-----------------------------------------------------------------------------------------------

5.2 Get deployments details

deployments_details = client.deployments.get_details(deployment_id)
print(json.dumps(deployments_details, indent=2))

List deployments

client.deployments.list()

5.3 Score deployed model

Let's plot two digits. Action: Please install matplotlib, numpy

import wget

dataset_filename = "mnist.npz"

if not os.path.isfile(dataset_filename):
    dataset_filename = wget.download("https://s3.amazonaws.com/img-datasets/mnist.npz")

import numpy as np

mnist_dataset = np.load(dataset_filename)
x_test = mnist_dataset["x_test"]

%matplotlib inline
import matplotlib.pyplot as plt

for i, image in enumerate([x_test[0], x_test[1]]):
    plt.subplot(2, 2, i + 1)
    plt.axis("off")
    plt.imshow(image, cmap=plt.cm.gray_r, interpolation="nearest")

Our input node expects to get data with shape (784,) so we need to reshape our two digits.

image_1 = x_test[0].ravel() / 255
image_2 = x_test[1].ravel() / 255

Prepare scoring payload and score.

scoring_payload = {
    client.deployments.ScoringMetaNames.INPUT_DATA: [
        {"values": [image_1.tolist(), image_2.tolist()]}
    ]
}
scores = client.deployments.score(deployment_id, meta_props=scoring_payload)
print("Scoring result:\n" + json.dumps(scores, indent=2))

Scoring result:
{
  "predictions": [
    {
      "fields": [
        "prediction",
        "prediction_classes",
        "probability"
      ],
      "id": "dense_2",
      "values": [
        [
          [
            1.2173110519597685e-08,
            5.5019970801595264e-08,
            7.814644050085917e-06,
            5.863983005838236e-06,
            3.238853629738969e-11,
            1.8598962014948484e-07,
            1.489450815288229e-12,
            0.9999855756759644,
            5.194610963599189e-08,
            4.473810975014203e-07
          ],
          7,
          [
            1.2173110519597685e-08,
            5.5019970801595264e-08,
            7.814644050085917e-06,
            5.863983005838236e-06,
            3.238853629738969e-11,
            1.8598962014948484e-07,
            1.489450815288229e-12,
            0.9999855756759644,
            5.194610963599189e-08,
            4.473810975014203e-07
          ]
        ],
        [
          [
            7.956906244999118e-08,
            0.0008763494552113116,
            0.9991222023963928,
            6.770528102606477e-07,
            6.7667859163234034e-12,
            6.170370170366368e-07,
            9.471719408793433e-09,
            1.5386401042150055e-09,
            6.930226170709375e-09,
            7.317644754034625e-12
          ],
          2,
          [
            7.956906244999118e-08,
            0.0008763494552113116,
            0.9991222023963928,
            6.770528102606477e-07,
            6.7667859163234034e-12,
            6.170370170366368e-07,
            9.471719408793433e-09,
            1.5386401042150055e-09,
            6.930226170709375e-09,
            7.317644754034625e-12
          ]
        ]
      ]
    }
  ]
}

6. Clean up

If you want to clean up all created assets:

• experiments

• trainings

• pipelines

• model definitions

• models

• functions

• deployments

please follow up this sample notebook.

7. Summary and next steps

You successfully completed this notebook! You learned how to use ibm-watsonx-ai to run experiments.

Check out our Online Documentation for more samples, tutorials, documentation, how-tos, and blog posts.

Author

Jan Sołtysik, Intern at watsonx.ai.

Rafał Chrzanowski, Software Engineer Intern at watsonx.ai.

Copyright © 2020-2025 IBM. This notebook and its source code are released under the terms of the MIT License.