Skip to content

Latest commit

 

History

History
250 lines (192 loc) · 7.4 KB

README_tools.md

File metadata and controls

250 lines (192 loc) · 7.4 KB
Raw

MLflow Export Import - Tools

Overview

Model signature tools

These tools are useful for upgrading Databricks Workspace Registry models to Unity Catalog Registry models which require a model signature.

Model signature scripts:

Test usage:

MLflow documentation:

Get model signature

Overview

Run examples

get-model-signature \
  --model-uri models:/Sklearn_Wine/12 

get-model-signature \
  --model-uri runs:/73ab168e5775409fa3595157a415bb62/model

{
  "inputs": [
    {
      "type": "double",
      "name": "fixed_acidity",
      "required": true
    },
    {
    . . .
}

Options

get_model_signature --help

  Get the signature of an MLflow model.

Options:
  --model-uri TEXT    Model URI such as 'models:/my-model/3' or
                      'runs:/73ab168e5775409fa3595157a415bb62/my-model'.
                      [required]
  --output-file TEXT  Output file.

Set model signature

Overview

model artifacts located in the model registry and represented by models:/ URI schemes are not compatible with this API

  • For OSS MLflow, if you add a model signature to a run model, it will automatically update any model version that was created from that run. AFAIK this is not documented.
  • Source: set_model_signature.py.

Run examples

set-model-signature \
  --model-uri runs:/73ab168e5775409fa3595157a415bb62/model \
  --input-file training_samples.csv \
  --output-file predictions_samples.csv

set-model-signature \
  --model-uri file:/run/model \
  --input-file training_samples.csv \
  --output-file predictions_samples.csv

Options

set-model-signature --help \

  Set the signature of an MLflow model.

Options:
  --model-uri TEXT               Model URI such as 'runs:/73ab168e5775409fa359
                                 5157a415bb62/my_model' or
                                 'file:/my_mlflow_model/.  Per MLflow
                                 documentation 'models:/' scheme is not
                                 supported.  [required]
  --input-file TEXT              Input CSV file with training data samples.
                                 [required]
  --output-file TEXT             Output CSV file with prediction data samples.
  --overwrite-signature BOOLEAN  Overwrite existing model signature.

List model versions without a model signature

Overview

Run examples

list-model-versions-without-signatures \
  --output-file model-versions-without-signatures.csv

list-model-versions-without-signatures \
  --filter "name like 'sklearn%' \
  --output-file model-versions-without-signatures.csv

+------------------+-----------+----------------------------------+---------+
| model            |   version | run_id                           | error   |
|------------------|-----------|----------------------------------|---------|
| sklearn_iris     |         1 | de99a1259e7e47ebb3c701d13a7680b0 |         |
| sklearn_wine     |        27 | 4f6355ff885e4e158cdfbc5e77b47d33 |         |
| sklearn_wine     |        24 | 80a7b81527c2462abd168cb0f549a33b |         |
| sklearn_wine     |        18 | f8fca475b2cc4995b75b5a51236e1251 |         |
| tensorflow_mnist |        11 | d8d008e022504dc78cbc9799a9e5d3b3 |         |
| xgboost_wine     |         1 | 53292dc36dca423ebfb8a33667143a74 |         |
+------------------+-----------+----------------------------------+---------+

Options

list-model-versions-without-signatures --help

  List model versions without a model signature.

Options:
  --filter TEXT       For OSS MLflow this is a filter for
                      search_model_versions(), for Databricks it is for
                      search_registered_models() due to Databricks MLflow
                      search limitations.
  --output-file TEXT  Output file.

Download notebook with revision

This tool downloads a notebook with a specific revision.

Note that the parameter revision_timestamp which represents the revision ID to the API endpoint workspace/export is not publicly documented.

Usage
download-notebook --help

Options:
  --output-dir TEXT        Output directory.  [required]
  --notebook TEXT          Notebook path.  [required]
  --revision TEXT          Notebook revision. If not specified will download
                           the latest revision.
  --notebook-formats TEXT  Databricks notebook formats. Values are SOURCE,
                           HTML, JUPYTER or DBC (comma seperated).  [default:
                           SOURCE]
Run examples
download-notebook \
  --output-dir out \
  --notebook /Users/andre@mycompany.com/mlflow/02a_Sklearn_Train_Predict \
  --revision 12345 \
  --notebook-formats SOURCE,DBC

ls -c1 out

02a_Sklearn_Train_Predict.dbc
02a_Sklearn_Train_Predict.source

Call http_client - MLflow API or Databricks API

A convenience script to directly invoke either the MLflow API or Databricks API.

Usage
http-client --help

Options:
Options:
  --api TEXT          API: mlflow|databricks.
  --resource TEXT     API resource such as 'experiments/search'.  [required]
  --method TEXT       HTTP method: GET|POST|PUT|PATCH|DELETE.
  --params TEXT       HTTP GET query parameters as JSON.
  --data TEXT         HTTP request entity body as JSON (for POST, PUT and
                      PATCH).
  --output-file TEXT  Output file.
HTTP GET example
export MLFLOW_TRACKING_URI=http://localhost:5000

http-client \
  --resource experiments/list \
  --output-file experiments.json
HTTP GET example - Databricks API
http-client \
  --api databricks \
  --resource clusters/list \
  --output-file experiments.json
HTTP POST example
export MLFLOW_TRACKING_URI=http://localhost:5000

http-client \
  --resource experiments/create \
  --data '{"name": "my_experiment"}'

Workflow API

  • README.md
  • The WorkflowApiClient is a Python wrapper around the Databricks REST API to execute job runs in a synchronous polling manner.
  • Although a generic tool, in terms of mlflow-export-import, its main use is for testing Databricks notebook jobs.