Model Explainer Example

Explainer for Iris model with Poetry-defined Environment

Prerequisites

• A kubernetes cluster with kubectl configured

• poetry

• rclone

• curl

Setup Seldon Core

Use the setup notebook to Setup Cluster with Ambassador Ingress and Install Seldon Core. Instructions also online.

We will assume that ambassador (or Istio) ingress is port-forwarded to localhost:8003

Setup MinIO

Use the provided notebook to install Minio in your cluster. Instructions also online.

We will assume that MinIO service is port-forwarded to localhost:8090

%%writefile rclone.conf
[s3]
type = s3
provider = minio
env_auth = false
access_key_id = minioadmin
secret_access_key = minioadmin
endpoint = http://localhost:8090

Overwriting rclone.conf

%%writefile secret.yaml
apiVersion: v1
kind: Secret
metadata:
  name: seldon-rclone-secret
type: Opaque
stringData:
  RCLONE_CONFIG_S3_TYPE: s3
  RCLONE_CONFIG_S3_PROVIDER: minio
  RCLONE_CONFIG_S3_ENV_AUTH: "false"
  RCLONE_CONFIG_S3_ACCESS_KEY_ID: minioadmin
  RCLONE_CONFIG_S3_SECRET_ACCESS_KEY: minioadmin
  RCLONE_CONFIG_S3_ENDPOINT: http://minio.minio-system.svc.cluster.local:9000

Overwriting secret.yaml

!kubectl apply -f secret.yaml

secret/seldon-rclone-secret configured

Poetry

We will use poetry.lock to fully define the explainer environment. Install poetry following official documentation. Usually this goes down to

curl -sSL https://install.python-poetry.org | python3 - --version 1.1.15

Deploy Iris Model

%%writefile iris.yaml
apiVersion: machinelearning.seldon.io/v1alpha2
kind: SeldonDeployment
metadata:
  name: iris
spec:
  predictors:
  - name: default
    replicas: 1
    graph:
      name: classifier
      implementation: SKLEARN_SERVER
      modelUri: gs://seldon-models/v1.20.0-dev/sklearn/iris

Overwriting iris.yaml

!kubectl apply -f iris.yaml

seldondeployment.machinelearning.seldon.io/iris configured

!kubectl rollout status deploy/$(kubectl get deploy -l seldon-deployment-id=iris -o jsonpath='{.items[0].metadata.name}')

Waiting for deployment "iris-default-0-classifier" rollout to finish: 1 old replicas are pending termination...
Waiting for deployment "iris-default-0-classifier" rollout to finish: 1 old replicas are pending termination...
deployment "iris-default-0-classifier" successfully rolled out

%%bash
curl -s -X POST -H 'Content-Type: application/json' \
    -d '{"data":{"ndarray":[[5.964, 4.006, 2.081, 1.031]]}}' \
    http://localhost:8003/seldon/seldon/iris/api/v1.0/predictions  | jq .

{
  "data": {
    "names": [
      "t:0",
      "t:1",
      "t:2"
    ],
    "ndarray": [
      [
        0.9548873249364059,
        0.04505474761562512,
        5.7927447968953825e-05
      ]
    ]
  },
  "meta": {
    "requestPath": {
      "classifier": "seldonio/sklearnserver:1.15.0-dev"
    }
  }
}

Train Explainer

Prepare Training Environment

We are going to use pyproject.toml and poetry.lock files from Alibi Explain Server. This will allow us to create environment that will match the runtime one.

%%bash
cp ../../../components/alibi-explain-server/pyproject.toml .
cp ../../../components/alibi-explain-server/poetry.lock .

conda create --yes --prefix ./venv python=3.7.10

Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): ...working... done
Solving environment: ...working... done

## Package Plan ##

  environment location: /home/rskolasinski/work/seldon-core/examples/explainers/iris-explainer-poetry/venv

  added / updated specs:
    - conda-ecosystem-user-package-isolation
    - python=3.7.10


The following NEW packages will be INSTALLED:

  _libgcc_mutex      conda-forge/linux-64::_libgcc_mutex-0.1-conda_forge
  _openmp_mutex      conda-forge/linux-64::_openmp_mutex-4.5-2_gnu
  ca-certificates    conda-forge/linux-64::ca-certificates-2022.5.18.1-ha878542_0
  conda-ecosystem-u~ conda-forge/linux-64::conda-ecosystem-user-package-isolation-1.0-ha770c72_1
  ld_impl_linux-64   conda-forge/linux-64::ld_impl_linux-64-2.36.1-hea4e1c9_2
  libffi             conda-forge/linux-64::libffi-3.4.2-h7f98852_5
  libgcc-ng          conda-forge/linux-64::libgcc-ng-12.1.0-h8d9b700_16
  libgomp            conda-forge/linux-64::libgomp-12.1.0-h8d9b700_16
  libnsl             conda-forge/linux-64::libnsl-2.0.0-h7f98852_0
  libstdcxx-ng       conda-forge/linux-64::libstdcxx-ng-12.1.0-ha89aaad_16
  libzlib            conda-forge/linux-64::libzlib-1.2.11-h166bdaf_1014
  ncurses            conda-forge/linux-64::ncurses-6.3-h27087fc_1
  openssl            conda-forge/linux-64::openssl-3.0.3-h166bdaf_0
  pip                conda-forge/noarch::pip-22.1.1-pyhd8ed1ab_0
  python             conda-forge/linux-64::python-3.7.10-hf930737_104_cpython
  python_abi         conda-forge/linux-64::python_abi-3.7-2_cp37m
  readline           conda-forge/linux-64::readline-8.1-h46c0cb4_0
  setuptools         conda-forge/linux-64::setuptools-62.3.2-py37h89c1867_0
  sqlite             conda-forge/linux-64::sqlite-3.38.5-h4ff8645_0
  tk                 conda-forge/linux-64::tk-8.6.12-h27826a3_0
  wheel              conda-forge/noarch::wheel-0.37.1-pyhd8ed1ab_0
  xz                 conda-forge/linux-64::xz-5.2.5-h516909a_1
  zlib               conda-forge/linux-64::zlib-1.2.11-h166bdaf_1014


Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done
#
# To activate this environment, use
#
#     $ conda activate /home/rskolasinski/work/seldon-core/examples/explainers/iris-explainer-poetry/venv
#
# To deactivate an active environment, use
#
#     $ conda deactivate

%%bash
source ~/miniconda3/etc/profile.d/conda.sh
conda activate ./venv
poetry install

Installing dependencies from lock file

Package operations: 138 installs, 0 updates, 0 removals

  • Installing certifi (2021.10.8)
  • Installing charset-normalizer (2.0.11)
  • Installing idna (3.3)
  • Installing pyasn1 (0.4.8)
  • Installing pycparser (2.21)
  • Installing typing-extensions (4.0.1)
  • Installing urllib3 (1.26.8)
  • Installing zipp (3.7.0)
  • Installing cachetools (5.0.0)
  • Installing cffi (1.15.0)
  • Installing cymem (2.0.6)
  • Installing importlib-metadata (4.10.1)
  • Installing numpy (1.19.5)
  • Installing oauthlib (3.2.0)
  • Installing pyparsing (3.0.7)
  • Installing requests (2.27.1)
  • Installing rsa (4.7.2)
  • Installing six (1.16.0)
  • Installing pyasn1-modules (0.2.8)
  • Installing murmurhash (1.0.6)
  • Installing blis (0.7.5)
  • Installing catalogue (1.0.0)
  • Installing click (8.0.3)
  • Installing cryptography (36.0.1)
  • Installing filelock (3.4.2)
  • Installing google-auth (2.6.0)
  • Installing joblib (1.1.0)
  • Installing packaging (21.3)
  • Installing pillow (9.0.0)
  • Installing preshed (3.0.6)
  • Installing httplib2 (0.20.2)
  • Installing pyyaml (6.0)
  • Installing pyu2f (0.1.5)
  • Installing plac (1.1.3)
  • Installing requests-oauthlib (1.3.1)
  • Installing regex (2022.1.18)
  • Installing sniffio (1.2.0)
  • Installing srsly (1.0.5)
  • Installing tomli (1.2.3)
  • Installing tqdm (4.62.3)
  • Installing wasabi (0.9.0)
  • Installing absl-py (1.0.0)
  • Installing anyio (3.5.0)
  • Installing attrs (21.4.0)
  • Installing boto (2.49.0)
  • Installing cached-property (1.5.2)
  • Installing cycler (0.11.0)
  • Installing fasteners (0.17.3)
  • Installing fonttools (4.29.1)
  • Installing google-auth-oauthlib (0.4.6)
  • Installing google-reauth (0.1.1)
  • Installing grpcio (1.43.0)
  • Installing huggingface-hub (0.4.0)
  • Installing imageio (2.14.1)
  • Installing iniconfig (1.1.1)
  • Installing kiwisolver (1.3.2)
  • Installing llvmlite (0.38.0)
  • Installing markdown (3.3.6)
  • Installing networkx (2.6.3)
  • Installing oauth2client (4.1.3)
  • Installing pluggy (1.0.0)
  • Installing protobuf (3.19.4)
  • Installing py (1.11.0)
  • Installing python-dateutil (2.8.2)
  • Installing pyopenssl (22.0.0)
  • Installing pytz (2021.3)
  • Installing pywavelets (1.2.0)
  • Installing retry-decorator (1.1.1)
  • Installing sacremoses (0.0.47)
  • Installing scipy (1.7.3)
  • Installing setuptools-scm (6.4.2)
  • Installing spacy-lookups-data (0.3.2)
  • Installing tenacity (8.0.1)
  • Installing tensorboard-data-server (0.6.1)
  • Installing tensorboard-plugin-wit (1.8.1)
  • Installing thinc (7.4.5)
  • Installing threadpoolctl (3.1.0)
  • Installing tifffile (2021.11.2)
  • Installing tokenizers (0.11.4)
  • Installing toml (0.10.2)
  • Installing werkzeug (2.0.2)
  • Installing appdirs (1.4.4)
  • Installing argcomplete (2.0.0)
  • Installing astunparse (1.6.3)
  • Installing cloudpickle (2.0.0)
  • Installing crcmod (1.7)
  • Installing dill (0.3.4)
  • Installing gast (0.4.0)
  • Installing gcs-oauth2-boto-plugin (3.0)
  • Installing google-pasta (0.2.0)
  • Installing graphviz (0.19.1)
  • Installing h5py (3.6.0)
  • Installing flatbuffers (2.0)
  • Installing keras-preprocessing (1.1.2)
  • Installing keras (2.7.0)
  • Installing libclang (13.0.0)
  • Installing google-apitools (0.5.32)
  • Installing matplotlib (3.5.1)
  • Installing monotonic (1.6)
  • Installing mypy-extensions (0.4.3)
  • Installing numba (0.55.0)
  • Installing opt-einsum (3.3.0)
  • Installing pandas (1.1.5)
  • Installing pathspec (0.9.0)
  • Installing plotly (5.5.0)
  • Installing ptable (0.9.2)
  • Installing pydantic (1.9.0)
  • Installing pytest (6.2.5)
  • Installing scikit-image (0.19.1)
  • Installing scikit-learn (1.0.2)
  • Installing slicer (0.0.7)
  • Installing spacy (2.3.7)
  • Installing starlette (0.17.1)
  • Installing tensorboard (2.8.0)
  • Installing tensorflow-estimator (2.7.0)
  • Installing tensorflow-io-gcs-filesystem (0.23.1)
  • Installing termcolor (1.1.0)
  • Installing tornado (6.1)
  • Installing transformers (4.16.2)
  • Installing typed-ast (1.4.3)
  • Installing wrapt (1.13.3)
  • Installing alibi (0.7.0)
  • Installing black (21.7b0)
  • Installing catboost (1.0.4)
  • Installing fastapi (0.73.0)
  • Installing grpcio-tools (1.31.0)
  • Installing gsutil (5.5)
  • Installing isort (5.9.0)
  • Installing lightgbm (3.3.2)
  • Installing mypy (0.910)
  • Installing mypy-protobuf (1.22)
  • Installing pip-licenses (3.5.3)
  • Installing pytest-tornasync (0.6.0.post2)
  • Installing requests-mock (1.9.3)
  • Installing shap (0.40.0 429fb3e)
  • Installing tensorflow (2.7.0)
  • Installing types-requests (2.26.0)
  • Installing xgboost (1.5.2)

Prepare Training Script

%%writefile train.py
import numpy as np
from sklearn.datasets import load_iris
from alibi.explainers import AnchorTabular

import requests


dataset = load_iris()
feature_names = dataset.feature_names
iris_data = dataset.data

model_url = "http://localhost:8003/seldon/seldon/iris/api/v1.0/predictions"


def predict_fn(X):
    data = {"data": {"ndarray": X.tolist()}}
    r = requests.post(model_url, json={"data": {"ndarray": [[1, 2, 3, 4]]}})
    return np.array(r.json()["data"]["ndarray"])


explainer = AnchorTabular(predict_fn, feature_names)
explainer.fit(iris_data, disc_perc=(25, 50, 75))

explainer.save("./explainer/")

Overwriting train.py

%%bash
unset MPLBACKEND # required as we call the script from Jupyter Lab in this demo
./venv/bin/python3 train.py

IPython could not be loaded!

!tree explainer/

explainer/
├── explainer.dill
└── meta.dill

0 directories, 2 files

Save and deploy Explainer

!rclone --config="rclone.conf" copy explainer/ s3:explainers/iris/

%%writefile iris-with-explainer.yaml
apiVersion: machinelearning.seldon.io/v1alpha2
kind: SeldonDeployment
metadata:
  name: iris
spec:
  predictors:
  - name: default
    replicas: 1
    graph:
      name: classifier
      implementation: SKLEARN_SERVER
      modelUri: gs://seldon-models/v1.20.0-dev/sklearn/iris
    explainer:
      type: AnchorTabular
      modelUri: s3:explainers/iris/
      envSecretRefName: seldon-rclone-secret
      replicas: 1            

Overwriting iris-with-explainer.yaml

!kubectl apply -f iris-with-explainer.yaml

seldondeployment.machinelearning.seldon.io/iris configured

!kubectl rollout status deploy/$(kubectl get deploy -l seldon-deployment-id=iris -o jsonpath='{.items[0].metadata.name}')
!kubectl rollout status deploy/$(kubectl get deploy -l seldon-deployment-id=iris -o jsonpath='{.items[1].metadata.name}')

Waiting for deployment "iris-default-0-classifier" rollout to finish: 1 old replicas are pending termination...
Waiting for deployment "iris-default-0-classifier" rollout to finish: 1 old replicas are pending termination...
deployment "iris-default-0-classifier" successfully rolled out
deployment "iris-default-explainer" successfully rolled out

Test Deployed explainer

import numpy as np
import requests

model_url = (
    "http://localhost:8003/seldon/seldon/iris-explainer/default/api/v1.0/explain"
)


def explain_fn(X):
    data = {"data": {"ndarray": X.tolist()}}
    r = requests.post(model_url, json={"data": {"ndarray": [[1, 2, 3, 4]]}})
    return r.json()

explanation = explain_fn(np.array([[5.964, 4.006, 2.081, 1.031]]))

print("Anchor: %s" % (" AND ".join(explanation["data"]["anchor"])))
print("Precision: %.2f" % explanation["data"]["precision"])
print("Coverage: %.2f" % explanation["data"]["coverage"])

Anchor: petal width (cm) > 1.80 AND sepal width (cm) <= 2.80
Precision: 0.97
Coverage: 0.08