Model Optimization

mikova — Mon, 24 Mar 2025 21:59:41 GMT

Download this notebook from: semantic-link-labs/notebooks/Model Optimization.ipynb at main · microsoft/semantic-link-labs · GitHub

Install the latest .whl package

Check here to see the latest version.

%pip install semantic-link-labs

Import the library

import sempy_labs as labs
from sempy_labs import lakehouse as lake
from sempy_labs import directlake
import sempy_labs.report as rep

dataset_name = ''
workspace_name = None

Vertipaq Analyzer

labs.vertipaq_analyzer(dataset=dataset_name, workspace=workspace_name)

Export the Vertipaq Analyzer results to a .zip file in your lakehouse

labs.vertipaq_analyzer(dataset=dataset_name, workspace=workspace_name, export='zip')

Export the Vertipaq Analyzer results to append to delta tables in your lakehouse.

labs.vertipaq_analyzer(dataset=dataset_name, workspace=workspace_name, export='table')

Visualize the contents of an exported Vertipaq Analzyer .zip file.

labs.import_vertipaq_analyzer(folder_path='', file_name='')

Best Practice Analyzer

This runs the standard rules for semantic models posted on Microsoft's GitHub.

labs.run_model_bpa(dataset=dataset_name, workspace=workspace_name)

This runs the Best Practice Analyzer and exports the results to the 'modelbparesults' delta table in your Fabric lakehouse.

labs.run_model_bpa(dataset=dataset_name, workspace=workspace_name, export=True)

This runs the Best Practice Analyzer with the rules translated into Italian (can enter any language in the 'language' parameter).

labs.run_model_bpa(dataset=dataset_name, workspace=workspace_name, language='italian')

Note: For analyzing model BPA results at scale, see the Best Practice Analyzer Report notebook (link below).

Best Practice Analyzer Report

Run BPA using your own best practice rules

import sempy
sempy.fabric._client._utils._init_analysis_services()
import Microsoft.AnalysisServices.Tabular as TOM
import pandas as pd

dataset_name = ''
workspace_name = ''

rules = pd.DataFrame(
    [
        (
            "Performance",
            "Table",
            "Warning",
            "Rule name...",
            lambda obj, tom: tom.is_calculated_table(table_name=obj.Name),
            'Rule description...',
            '',
        ),
        (
            "Performance",
            "Column",
            "Warning",
            "Do not use floating point data types",
            lambda obj, tom: obj.DataType == TOM.DataType.Double,
            'The "Double" floating point data type should be avoided, as it can result in unpredictable roundoff errors and decreased performance in certain scenarios. Use "Int64" or "Decimal" where appropriate (but note that "Decimal" is limited to 4 digits after the decimal sign).',
        )
    ],
    columns=[
            "Category",
            "Scope",
            "Severity",
            "Rule Name",
            "Expression",
            "Description",
            "URL",
        ],
)

labs.run_model_bpa(dataset=dataset_name, workspace=workspace_name, rules=rules)

Translate a semantic model's metadata

labs.translate_semantic_model(dataset=dataset_name, workspace=workspace_name, languages=['italian', 'japanese', 'hindi'], exclude_characters='_')

Direct Lake

Check if any lakehouse tables will hit the Direct Lake guardrails.

lake.get_lakehouse_tables(lakehouse=None, workspace=None, extended=True, count_rows=False)

lake.get_lakehouse_tables(lakehouse=None, workspace=None, extended=True, count_rows=False, export=True)

Check if any tables in a Direct Lake semantic model will fall back to DirectQuery.

directlake.check_fallback_reason(dataset=dataset_name, workspace=workspace_name)

OPTIMIZE your lakehouse delta tables.

lake.optimize_lakehouse_tables(tables=['', ''], lakehouse=None, workspace=None)

Refresh/reframe your Direct Lake semantic model and restore the columns which were in memory prior to the refresh.

directlake.warm_direct_lake_cache_isresident(dataset=dataset_name, workspace=workspace_name)

Ensure a warm cache for your users by putting the columns of a Direct Lake semantic model into memory based on the contents of a perspective.

Perspectives can be created either in Tabular Editor 3 or in Tabular Editor 2 using the Perspective Editor.

directlake.warm_direct_lake_cache_perspective(dataset=dataset_name, workspace=workspace_name, perspective='', add_dependencies=True)

https%3A%2F%2Fgithub.com%2Fmicrosoft%2Fsemantic-link-labs%2Fblob%2Fmain%2Fnotebooks%2FModel%2520Optimization.ipynb

topic Model Optimization in Notebook Gallery