Skip to content

Commit

Permalink
wip: use explain api for exclude fields
Browse files Browse the repository at this point in the history
  • Loading branch information
@alvarezmelissa87
alvarezmelissa87 committed Jan 6, 2020
1 parent 5cfd1c9 commit 24bdeb3
Show file tree
Hide file tree
Showing 3 changed files with 139 additions and 101 deletions.
230 changes: 130 additions & 100 deletions ...ics/pages/analytics_management/components/create_analytics_form/create_analytics_form.tsx
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ import { IndexPattern, INDEX_PATTERN_ILLEGAL_CHARACTERS } from 'ui/index_pattern
import { ES_FIELD_TYPES } from '../../../../../../../../../../../src/plugins/data/public';
import { ml } from '../../../../../services/ml_api_service';
import { Field, EVENT_RATE_FIELD_ID } from '../../../../../../../common/types/fields';

import { newJobCapsService } from '../../../../../services/new_job_capabilities_service';
import { useKibanaContext } from '../../../../../contexts/kibana';
import { CreateAnalyticsFormProps } from '../../hooks/use_create_analytics_form';
Expand All @@ -45,24 +44,20 @@ import {
indexPatterns,
} from '../../../../../../../../../../../src/plugins/data/public';

const NUMERICAL_FIELD_TYPES = new Set([
const CATEGORICAL_NUMERICAL_TYPES = new Set([
ES_FIELD_TYPES.LONG,
ES_FIELD_TYPES.INTEGER,
ES_FIELD_TYPES.SHORT,
ES_FIELD_TYPES.BYTE,
]);

const NUMERICAL_TYPES = new Set([
ES_FIELD_TYPES.DOUBLE,
ES_FIELD_TYPES.FLOAT,
ES_FIELD_TYPES.HALF_FLOAT,
ES_FIELD_TYPES.SCALED_FLOAT,
]);

const SUPPORTED_ANALYSIS_FIELD_TYPES = new Set([
ES_FIELD_TYPES.BOOLEAN,
ES_FIELD_TYPES.TEXT,
ES_FIELD_TYPES.KEYWORD,
ES_FIELD_TYPES.IP,
]);

// List of system fields we want to ignore for the numeric field check.
const OMIT_FIELDS: string[] = ['_source', '_type', '_index', '_id', '_version', '_score'];

Expand All @@ -84,15 +79,19 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
destinationIndexPatternTitleExists,
excludes,
excludesOptions,
fieldOptionsFetchFail,
jobId,
jobIdEmpty,
jobIdExists,
jobIdValid,
jobIdInvalidMaxLength,
jobType,
loadingDepVarOptions,
loadingFieldOptions,
modelMemoryLimit,
modelMemoryLimitUnitValid,
previousJobType,
previousSourceIndex,
sourceIndex,
sourceIndexNameEmpty,
sourceIndexNameValid,
Expand All @@ -102,6 +101,9 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
} = form;
const characterList = indexPatterns.ILLEGAL_CHARACTERS_VISIBLE.join(', ');

const isJobTypeWithDepVar =
jobType === JOB_TYPES.REGRESSION || jobType === JOB_TYPES.CLASSIFICATION;

// Find out if index pattern contain numeric fields. Provides a hint in the form
// that an analytics jobs is not able to identify outliers if there are no numeric fields present.
const validateSourceIndexFields = async () => {
Expand All @@ -124,27 +126,18 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
}
};

// Regression and Classification support numeric, boolean, text, keyword, and ip types
// Outlier detection supports numeric or boolean types
const shouldAddAsAnalyzedFieldsOption = (field: Field) => {
if (field.id === EVENT_RATE_FIELD_ID) return false;

const isNumerical = NUMERICAL_FIELD_TYPES.has(field.type);

if (jobType === JOB_TYPES.OUTLIER_DETECTION)
return isNumerical || field.type === ES_FIELD_TYPES.BOOLEAN;
if (jobType === JOB_TYPES.REGRESSION || jobType === JOB_TYPES.CLASSIFICATION)
return isNumerical || SUPPORTED_ANALYSIS_FIELD_TYPES.has(field.type);
};

// Regression supports numeric fields. Classification supports numeric and boolean.
// Regression supports numeric fields. Classification supports categorical numeric and boolean.
const shouldAddAsDepVarOption = (field: Field) => {
if (field.id === EVENT_RATE_FIELD_ID) return false;

const isNumerical = NUMERICAL_FIELD_TYPES.has(field.type);
const isSupportedByClassification = isNumerical || field.type === ES_FIELD_TYPES.BOOLEAN;
const isCategoricalNumerical = CATEGORICAL_NUMERICAL_TYPES.has(field.type);

const isSupportedByClassification =
isCategoricalNumerical || field.type === ES_FIELD_TYPES.BOOLEAN;

if (jobType === JOB_TYPES.REGRESSION) return isNumerical;
if (jobType === JOB_TYPES.REGRESSION) {
return isCategoricalNumerical || NUMERICAL_TYPES.has(field.type);
}
if (jobType === JOB_TYPES.CLASSIFICATION) return isSupportedByClassification;
};

Expand All @@ -170,17 +163,53 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
}
};

const debouncedMmlEstimateLoad = debounce(async () => {
const debouncedGetExplainData = debounce(async () => {
// Reset if sourceIndex or jobType changes (jobType requires dependent_variable to be set -
// which won't be the case if switching from outlier detection)
if (previousSourceIndex !== sourceIndex || previousJobType !== jobType) {
setFormState({
loadingFieldOptions: true,
// excludesOptions: [],
// excludes: [],
});
}

try {
const jobConfig = getJobConfigFromFormState(form);
delete jobConfig.dest;
delete jobConfig.model_memory_limit;
delete jobConfig.analyzed_fields;
const resp = await ml.dataFrameAnalytics.explainDataFrameAnalytics(jobConfig);
setFormState({
modelMemoryLimit: resp.memory_estimation?.expected_memory_without_disk,
});

// If sourceIndex has changed load analysis field options again
if (previousSourceIndex !== sourceIndex) {
const analyzedFieldsOptions: Array<{ label: string }> = [];

if (resp.field_selection) {
resp.field_selection.forEach((selectedField: any) => {
// TODO: update type
if (selectedField.is_included === true && selectedField.name !== dependentVariable) {
analyzedFieldsOptions.push({ label: selectedField.name });
}
});
}

setFormState({
modelMemoryLimit: resp.memory_estimation?.expected_memory_without_disk,
excludesOptions: analyzedFieldsOptions,
excludes: [],
loadingFieldOptions: false,
dependentVariableFetchFail: false,
});
} else {
setFormState({
modelMemoryLimit: resp.memory_estimation?.expected_memory_without_disk,
});
}
} catch (e) {
setFormState({
fieldOptionsFetchFail: true,
loadingFieldOptions: false,
modelMemoryLimit:
jobType !== undefined
? DEFAULT_MODEL_MEMORY_LIMIT[jobType]
Expand All @@ -189,57 +218,40 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
}
}, 500);

const loadFieldOptions = async () => {
const loadDepVarOptions = async () => {
setFormState({
loadingFieldOptions: true,
loadingDepVarOptions: true,
// clear when the source index changes
dependentVariable: '',
excludes: [],
// Reset outlier detection sourceIndex checks to default values if we've switched to regression
sourceIndexFieldsCheckFailed: false,
sourceIndexContainsNumericalFields: true,
});
try {
const indexPattern: IndexPattern = await kibanaContext.indexPatterns.get(
indexPatternsMap[sourceIndex].value
);
const shouldPrepDepVarOptions =
jobType === JOB_TYPES.REGRESSION || jobType === JOB_TYPES.CLASSIFICATION;

if (indexPattern !== undefined) {
await newJobCapsService.initializeFromIndexPattern(indexPattern);
// Get fields and filter for supported types for job type
const { fields } = newJobCapsService;

const depVarOptions: Array<{ label: string }> = [];
const analyzedFieldsOptions: Array<{ label: string }> = [];

fields.forEach((field: Field) => {
if (shouldPrepDepVarOptions && shouldAddAsDepVarOption(field)) {
if (shouldAddAsDepVarOption(field)) {
depVarOptions.push({ label: field.id });
}
if (shouldAddAsAnalyzedFieldsOption(field)) {
analyzedFieldsOptions.push({ label: field.id });
}
});

if (shouldPrepDepVarOptions) {
setFormState({
dependentVariableOptions: depVarOptions,
excludesOptions: analyzedFieldsOptions,
loadingFieldOptions: false,
dependentVariableFetchFail: false,
});
} else {
setFormState({
excludesOptions: analyzedFieldsOptions,
loadingFieldOptions: false,
dependentVariableFetchFail: false,
});
}
setFormState({
dependentVariableOptions: depVarOptions,
loadingDepVarOptions: false,
dependentVariableFetchFail: false,
});
}
} catch (e) {
setFormState({ loadingFieldOptions: false, dependentVariableFetchFail: true });
setFormState({ loadingDepVarOptions: false, dependentVariableFetchFail: true });
}
};

Expand Down Expand Up @@ -272,9 +284,10 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
};

useEffect(() => {
if (jobType !== undefined && sourceIndexNameEmpty === false) {
loadFieldOptions();
if (isJobTypeWithDepVar && sourceIndexNameEmpty === false) {
loadDepVarOptions();
}

if (jobType === JOB_TYPES.OUTLIER_DETECTION && sourceIndexNameEmpty === false) {
validateSourceIndexFields();
}
Expand All @@ -283,21 +296,18 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
useEffect(() => {
const hasBasicRequiredFields =
jobType !== undefined && sourceIndex !== '' && sourceIndexNameValid === true;
const jobTypesWithDepVar =
jobType === JOB_TYPES.REGRESSION || jobType === JOB_TYPES.CLASSIFICATION;

const hasRequiredAnalysisFields =
(jobTypesWithDepVar && dependentVariable !== '' && trainingPercent !== undefined) ||
jobType === JOB_TYPES.OUTLIER_DETECTION;
(isJobTypeWithDepVar && dependentVariable !== '') || jobType === JOB_TYPES.OUTLIER_DETECTION;

if (hasBasicRequiredFields && hasRequiredAnalysisFields) {
debouncedMmlEstimateLoad();
debouncedGetExplainData();
}

return () => {
debouncedMmlEstimateLoad.cancel();
debouncedGetExplainData.cancel();
};
}, [jobType, sourceIndex, dependentVariable, trainingPercent]);
}, [jobType, sourceIndex, sourceIndexNameEmpty, dependentVariable, trainingPercent]);

return (
<EuiForm className="mlDataFrameAnalyticsCreateForm">
Expand Down Expand Up @@ -414,9 +424,14 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
selectedOptions={
indexPatternsMap[sourceIndex] !== undefined ? [{ label: sourceIndex }] : []
}
onChange={selectedOptions =>
setFormState({ sourceIndex: selectedOptions[0].label || '' })
}
onChange={selectedOptions => {
setFormState({
excludes: [],
excludesOptions: [],
previousSourceIndex: sourceIndex,
sourceIndex: selectedOptions[0].label || '',
});
}}
isClearable={false}
data-test-subj="mlAnalyticsCreateJobFlyoutSourceIndexSelect"
/>
Expand Down Expand Up @@ -488,36 +503,6 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
data-test-subj="mlAnalyticsCreateJobFlyoutDestinationIndexInput"
/>
</EuiFormRow>
<EuiFormRow
label={i18n.translate('xpack.ml.dataframe.analytics.create.excludedFieldsLabel', {
defaultMessage: 'Excluded fields',
})}
helpText={i18n.translate('xpack.ml.dataframe.analytics.create.excludedFieldsHelpText', {
defaultMessage:
'Optionally select fields to be excluded from analysis. All other supported fields will be included',
})}
>
<EuiComboBox
aria-label={i18n.translate(
'xpack.ml.dataframe.analytics.create.excludesInputAriaLabel',
{
defaultMessage: 'Optional. Enter or select field to be excluded.',
}
)}
isDisabled={isJobCreated}
isLoading={loadingFieldOptions}
options={excludesOptions}
selectedOptions={excludes.map(field => ({
label: field,
}))}
onCreateOption={onCreateOption}
onChange={selectedOptions =>
setFormState({ excludes: selectedOptions.map(option => option.label) })
}
isClearable={true}
data-test-subj="mlAnalyticsCreateJobFlyoutExcludesSelect"
/>
</EuiFormRow>
{(jobType === JOB_TYPES.REGRESSION || jobType === JOB_TYPES.CLASSIFICATION) && (
<Fragment>
<EuiFormRow
Expand Down Expand Up @@ -566,12 +551,14 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
}
)}
isDisabled={isJobCreated}
isLoading={loadingFieldOptions}
isLoading={loadingDepVarOptions}
singleSelection={true}
options={dependentVariableOptions}
selectedOptions={dependentVariable ? [{ label: dependentVariable }] : []}
onChange={selectedOptions =>
setFormState({ dependentVariable: selectedOptions[0].label || '' })
setFormState({
dependentVariable: selectedOptions[0].label || '',
})
}
isClearable={false}
isInvalid={dependentVariable === ''}
Expand All @@ -598,6 +585,49 @@ export const CreateAnalyticsForm: FC<CreateAnalyticsFormProps> = ({ actions, sta
</EuiFormRow>
</Fragment>
)}
<EuiFormRow
label={i18n.translate('xpack.ml.dataframe.analytics.create.excludedFieldsLabel', {
defaultMessage: 'Excluded fields',
})}
helpText={i18n.translate('xpack.ml.dataframe.analytics.create.excludedFieldsHelpText', {
defaultMessage:
'Optionally select fields to be excluded from analysis. All other supported fields will be included',
})}
error={
excludesOptions.length === 0 &&
fieldOptionsFetchFail === false &&
!sourceIndexNameEmpty && [
i18n.translate(
'xpack.ml.dataframe.analytics.create.excludesOptionsNoSupportedFields',
{
defaultMessage:
'No supported analysis fields were found for this index pattern.',
}
),
]
}
>
<EuiComboBox
aria-label={i18n.translate(
'xpack.ml.dataframe.analytics.create.excludesInputAriaLabel',
{
defaultMessage: 'Optional. Enter or select field to be excluded.',
}
)}
isDisabled={isJobCreated}
isLoading={loadingFieldOptions}
options={excludesOptions}
selectedOptions={excludes.map(field => ({
label: field,
}))}
onCreateOption={onCreateOption}
onChange={selectedOptions =>
setFormState({ excludes: selectedOptions.map(option => option.label) })
}
isClearable={true}
data-test-subj="mlAnalyticsCreateJobFlyoutExcludesSelect"
/>
</EuiFormRow>
<EuiFormRow
label={i18n.translate('xpack.ml.dataframe.analytics.create.modelMemoryLimitLabel', {
defaultMessage: 'Model memory limit',
Expand Down
2 changes: 1 addition & 1 deletion ..._frame_analytics/pages/analytics_management/components/create_analytics_form/job_type.tsx
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,7 @@ export const JobType: FC<Props> = ({ type, setFormState }) => {
hasNoInitialSelection={true}
onChange={e => {
const value = e.target.value as AnalyticsJobType;
setFormState({ jobType: value });
setFormState({ previousJobType: type, jobType: value });
}}
data-test-subj="mlAnalyticsCreateJobFlyoutJobTypeSelect"
/>
Expand Down
Loading

0 comments on commit 24bdeb3

Please sign in to comment.