Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[ML] Data Frame Analytics: Fix feature importance #61761

Merged
merged 11 commits into from
Apr 4, 2020
Merged

[ML] Data Frame Analytics: Fix feature importance #61761

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
92c4083
[ML] Add num_top_feature_importance_values to analytics form.
walterra Mar 30, 2020
4956c29
[ML] Fix availability of feature_importance fields in field selection…
walterra Mar 30, 2020
4cc34ac
[ML] Fix default selected fields.
walterra Mar 30, 2020
784bb67
Merge branch 'master' into ml-fix-feature-importance
walterra Apr 1, 2020
24502fd
[ML] Fix num_top_feature_importance_values validation.
walterra Apr 1, 2020
0655843
[ML] Fix duplicate i18n id.
walterra Apr 1, 2020
f61359b
Merge branch 'master' into ml-fix-feature-importance
walterra Apr 3, 2020
b4e2b69
[ML] Fix form validation.
walterra Apr 3, 2020
238aa25
[ML] Type fix.
walterra Apr 3, 2020
3f08211
[ML] Fix form validation (again).
walterra Apr 3, 2020
6ad3355
[ML] Fix form validation (again).
walterra Apr 3, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions x-pack/plugins/ml/public/application/data_frame_analytics/common/analytics.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,7 @@ interface OutlierAnalysis {
interface Regression {
dependent_variable: string;
training_percent?: number;
num_top_feature_importance_values?: number;
prediction_field_name?: string;
}
export interface RegressionAnalysis {
Expand All @@ -44,6 +45,7 @@ interface Classification {
dependent_variable: string;
training_percent?: number;
num_top_classes?: string;
num_top_feature_importance_values?: number;
prediction_field_name?: string;
}
export interface ClassificationAnalysis {
Expand Down Expand Up @@ -186,6 +188,23 @@ export const getPredictionFieldName = (analysis: AnalysisConfig) => {
return predictionFieldName;
};

export const getNumTopFeatureImportanceValues = (analysis: AnalysisConfig) => {
// If undefined will be defaulted to dependent_variable when config is created
let numTopFeatureImportanceValues = 0;
if (
isRegressionAnalysis(analysis) &&
analysis.regression.num_top_feature_importance_values !== undefined
) {
numTopFeatureImportanceValues = analysis.regression.num_top_feature_importance_values;
} else if (
isClassificationAnalysis(analysis) &&
analysis.classification.num_top_feature_importance_values !== undefined
) {
numTopFeatureImportanceValues = analysis.classification.num_top_feature_importance_values;
}
return numTopFeatureImportanceValues;
};

export const getPredictedFieldName = (
resultsField: string,
analysis: AnalysisConfig,
Expand Down
25 changes: 21 additions & 4 deletions x-pack/plugins/ml/public/application/data_frame_analytics/common/fields.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,12 +7,13 @@
import { getNestedProperty } from '../../util/object_utils';
import {
DataFrameAnalyticsConfig,
getNumTopFeatureImportanceValues,
getPredictedFieldName,
getDependentVar,
getPredictionFieldName,
} from './analytics';
import { Field } from '../../../../common/types/fields';
import { ES_FIELD_TYPES } from '../../../../../../../src/plugins/data/public';
import { ES_FIELD_TYPES, KBN_FIELD_TYPES } from '../../../../../../../src/plugins/data/public';
import { newJobCapsService } from '../../services/new_job_capabilities_service';

export type EsId = string;
Expand Down Expand Up @@ -253,6 +254,7 @@ export const getDefaultFieldsFromJobCaps = (
const dependentVariable = getDependentVar(jobConfig.analysis);
const type = newJobCapsService.getFieldById(dependentVariable)?.type;
const predictionFieldName = getPredictionFieldName(jobConfig.analysis);
const numTopFeatureImportanceValues = getNumTopFeatureImportanceValues(jobConfig.analysis);
// default is 'ml'
const resultsField = jobConfig.dest.results_field;

Expand All @@ -261,6 +263,18 @@ export const getDefaultFieldsFromJobCaps = (
predictionFieldName ? predictionFieldName : defaultPredictionField
}`;

const featureImportanceFields = [];

if (numTopFeatureImportanceValues > 0) {
featureImportanceFields.push(
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
...fields.map(d => ({
id: `${resultsField}.feature_importance.${d.id}`,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This highlights that we need to use EuiDataGrid for classification and regression results, as I can't actually tell which columns are which!

image

name: `${resultsField}.feature_importance.${d.name}`,
type: KBN_FIELD_TYPES.NUMBER,
}))
);
}

const allFields: any = [
{
id: `${resultsField}.is_training`,
Expand All @@ -269,11 +283,14 @@ export const getDefaultFieldsFromJobCaps = (
},
{ id: predictedField, name: predictedField, type },
...fields,
...featureImportanceFields,
].sort(({ name: a }, { name: b }) => sortRegressionResultsFields(a, b, jobConfig));

let selectedFields = allFields
.slice(0, DEFAULT_REGRESSION_COLUMNS * 2)
.filter((field: any) => field.name === predictedField || !field.name.includes('.keyword'));
let selectedFields = allFields.filter(
(field: any) =>
field.name === predictedField ||
(!field.name.includes('.keyword') && !field.name.includes('.feature_importance.'))
);

if (selectedFields.length > DEFAULT_REGRESSION_COLUMNS) {
selectedFields = selectedFields.slice(0, DEFAULT_REGRESSION_COLUMNS);
Expand Down
6 changes: 6 additions & 0 deletions ...frame_analytics/pages/analytics_management/components/analytics_list/action_clone.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ describe('Analytics job clone action', () => {
classification: {
dependent_variable: 'y',
num_top_classes: 2,
num_top_feature_importance_values: 4,
prediction_field_name: 'y_prediction',
training_percent: 2,
randomize_seed: 6233212276062807000,
Expand Down Expand Up @@ -90,6 +91,7 @@ describe('Analytics job clone action', () => {
prediction_field_name: 'stab_prediction',
training_percent: 20,
randomize_seed: -2228827740028660200,
num_top_feature_importance_values: 4,
},
},
analyzed_fields: {
Expand Down Expand Up @@ -120,6 +122,7 @@ describe('Analytics job clone action', () => {
classification: {
dependent_variable: 'y',
num_top_classes: 2,
num_top_feature_importance_values: 4,
prediction_field_name: 'y_prediction',
training_percent: 2,
randomize_seed: 6233212276062807000,
Expand Down Expand Up @@ -188,6 +191,7 @@ describe('Analytics job clone action', () => {
prediction_field_name: 'stab_prediction',
training_percent: 20,
randomize_seed: -2228827740028660200,
num_top_feature_importance_values: 4,
},
},
analyzed_fields: {
Expand Down Expand Up @@ -218,6 +222,7 @@ describe('Analytics job clone action', () => {
dependent_variable: 'y',
training_percent: 71,
max_trees: 1500,
num_top_feature_importance_values: 4,
},
},
model_memory_limit: '400mb',
Expand All @@ -243,6 +248,7 @@ describe('Analytics job clone action', () => {
dependent_variable: 'y',
training_percent: 71,
maximum_number_trees: 1500,
num_top_feature_importance_values: 4,
},
},
model_memory_limit: '400mb',
Expand Down
4 changes: 4 additions & 0 deletions ...ata_frame_analytics/pages/analytics_management/components/analytics_list/action_clone.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,6 +97,8 @@ const getAnalyticsJobMeta = (config: CloneDataFrameAnalyticsConfig): AnalyticsJo
},
num_top_feature_importance_values: {
optional: true,
defaultValue: 0,
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
formKey: 'numTopFeatureImportanceValues',
},
class_assignment_objective: {
optional: true,
Expand Down Expand Up @@ -164,6 +166,8 @@ const getAnalyticsJobMeta = (config: CloneDataFrameAnalyticsConfig): AnalyticsJo
},
num_top_feature_importance_values: {
optional: true,
defaultValue: 0,
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
formKey: 'numTopFeatureImportanceValues',
},
randomize_seed: {
optional: true,
Expand Down
26 changes: 26 additions & 0 deletions ...ics/pages/analytics_management/components/create_analytics_form/create_analytics_form.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,6 +81,7 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
maxDistinctValuesError,
modelMemoryLimit,
modelMemoryLimitValidationResult,
numTopFeatureImportanceValues,
previousJobType,
previousSourceIndex,
sourceIndex,
Expand Down Expand Up @@ -643,6 +644,31 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
data-test-subj="mlAnalyticsCreateJobFlyoutTrainingPercentSlider"
/>
</EuiFormRow>
{/* num_top_feature_importance_values */}
<EuiFormRow
label={i18n.translate(
'xpack.ml.dataframe.analytics.create.numTopFeatureImportanceValuesLabel',
{
defaultMessage: 'Max. number of feature importance values per document',
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
}
)}
>
<EuiFieldText
disabled={false}
value={numTopFeatureImportanceValues}
onChange={e =>
setFormState({ numTopFeatureImportanceValues: parseInt(e.target.value, 10) })
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
}
aria-label={i18n.translate(
'xpack.ml.dataframe.analytics.create.numTopFeatureImportanceValuesInputAriaLabel',
{
defaultMessage: 'Max. number of feature importance values per document.',
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
}
)}
isInvalid={!destinationIndexNameEmpty && !destinationIndexNameValid}
data-test-subj="mlAnalyticsCreateJobFlyoutnumTopFeatureImportanceValuesInput"
/>
</EuiFormRow>
</Fragment>
)}
<EuiFormRow
Expand Down
4 changes: 4 additions & 0 deletions .../data_frame_analytics/pages/analytics_management/hooks/use_create_analytics_form/state.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,6 +69,7 @@ export interface State {
modelMemoryLimit: string | undefined;
modelMemoryLimitUnitValid: boolean;
modelMemoryLimitValidationResult: any;
numTopFeatureImportanceValues: number | undefined;
previousJobType: null | AnalyticsJobType;
previousSourceIndex: EsIndexName | undefined;
sourceIndex: EsIndexName;
Expand Down Expand Up @@ -123,6 +124,7 @@ export const getInitialState = (): State => ({
modelMemoryLimit: undefined,
modelMemoryLimitUnitValid: true,
modelMemoryLimitValidationResult: null,
numTopFeatureImportanceValues: 0,
peteharverson marked this conversation as resolved.
Show resolved Hide resolved
previousJobType: null,
previousSourceIndex: undefined,
sourceIndex: '',
Expand Down Expand Up @@ -182,6 +184,7 @@ export const getJobConfigFromFormState = (
jobConfig.analysis = {
[formState.jobType]: {
dependent_variable: formState.dependentVariable,
num_top_feature_importance_values: formState.numTopFeatureImportanceValues,
training_percent: formState.trainingPercent,
},
};
Expand Down Expand Up @@ -216,6 +219,7 @@ export function getCloneFormStateFromJobConfig(
const analysisConfig = analyticsJobConfig.analysis[jobType];

resultState.dependentVariable = analysisConfig.dependent_variable;
resultState.numTopFeatureImportanceValues = analysisConfig.num_top_feature_importance_values;
resultState.trainingPercent = analysisConfig.training_percent;
}

Expand Down