I have a trained onnx model that needs to be quantized to INT8. But I want my last fully connected layers are still in FP32 or FP16. So how can I choose specific layers to quantize (or not to quantize)?
PS when I was working with NNCF, I just use parametr ignored_scopes. Maybe is there something similar here at Workbench?
Use "ignored" parameter in configuration file to exclude nodes or operation types from optimization.
"algorithms": [
{
"name": "DefaultQuantization", // Optimization algorithm name
"params": {
"ignored": {
// List of nodes that are excluded from optimization
"scope": [
"<NODE_NAME>"
],
// List of types that are excluded from optimization
"operations": [
{
"type": "<NODE_TYPE>",
// Includes excluding by attributes
"attributes": {
"<NAME>": "<VALUE>" // Lists of values is not included
}
},
{
"type": "<NODE_TYPE>" // Excluding only by type
}
]
}
Related
Trying to setup schema markup for a simple math solver action with two fields. Let's say addition.
1+1=2
Here is Google's doc and example:
{
"#context": "https://schema.org",
"#type": ["MathSolver", "LearningResource"],
"name": "An awesome math solver",
"url": "https://www.mathdomain.com/",
"usageInfo": "https://www.mathdomain.com/privacy",
"inLanguage": "en",
"potentialAction": [{
"#type": "SolveMathAction",
"target": "https://mathdomain.com/solve?q={math_expression_string}",
"mathExpression-input": "required name=math_expression_string",
"eduQuestionType": ["Polynomial Equation","Derivative"]
}],
"learningResourceType": "Math solver"
}
How do we add multiple variables for two numbers?
return {
'#context': 'https://schema.org',
'#type': ['MathSolver', 'LearningResource'],
...
potentialAction: [
{
'#type': 'SolveMathAction',
target: `domain.com/?num1={num1}&num2={num2}`,
'mathExpression-input': 'required name=num1 name=num2',
eduQuestionType: ['addition', 'sum']
},
],
learningResourceType: 'Math solver'
};
Schema.org says about mathExpression (note: mathExpression-input doesnt seem to exist) but does fall under Thing > Intangible EntryPoint
A mathematical expression (e.g. 'x^2-3x=0') that may be solved for > a specific variable, simplified, or transformed. This can take many > formats, e.g. LaTeX, Ascii-Math, or math as you would write with a > keyboard.
But can this be setup for URL params to accept multiple fields within the mathExpression-input instead of a single math expression?
Can anyone give us a hint how to configure #semantic-release/release-notes-generator to take extra commit types (those additional to preset ones) while generating release-notes?
Our commits-analyzer config:
"#semantic-release/commit-analyzer",
{
"preset": "angular",
"releaseRules": [
{
"type": "refactor",
"release": "patch"
},
{
"type": "minfeat",
"release": "patch"
}
]
}
At this moment we have no extra configuration for #semantic-release/release-notes-generator, and surely it requires some as new types don't show up in changelog that is generated by #semantic-release/changelog after release-notes-generator runs
You will have to implement your own conventional-changelog preset in order to handle those new commit type.
For example the default angular preset handles only certain commit types: https://github.com/conventional-changelog/conventional-changelog/blob/e865af4df8d06795cebc7af09364ade19119e089/packages/conventional-changelog-angular/writer-opts.js#L36
Using segments, I am building an app that lets user enter a URL and check different metrics for it.
I see some significant differences between reports pulled with the API and report generated in the UI. Maybe I am misunderstanding something about segments.
For example, I have a segment designed to show only users that went on a specific page.
which map to a small fraction of my users:
When looking at the user age brackets, I see small numbers scattered across all categories.
Now to run the equivalent report in the API, I am using the payload below.
{
"reportRequests": [
{
"viewId": "#####",
"dateRanges": [
{ "startDate": "2017-03-01",
"endDate": "2017-04-27" }
],
"metrics": [
{"expression": "ga:pageviews"},
{"expression": "ga:sessions"},
{"expression": "ga:users"}
],
"segments": [
{
"dynamicSegment": {
"name": "Users of /apath/ofinterest/",
"userSegment": {
"segmentFilters": [
{
"simpleSegment": {
"orFiltersForSegment": {
"segmentFilterClauses": [
{
"dimensionFilter": {
"dimensionName": "ga:pagePath",
"operator": "EXACT",
"expressions": [
"/apath/ofinterest/"
]}}]}}}]}}}
],
"dimensions": [
{ "name": "ga:userAgeBracket" },
{ "name": "ga:segment" }
]
}
]
}
Which yield completely different results:
Only 2 age brackets, and weirdly, the same number of users in each (I tried with different time frame with the same behaviour).
Any ideas on what could be wrong? Could it be something in the settings of the segment? Related to "Sessions/User Include"?
Or could this warning below that I see in the UI have different impact in the UI and the API?
According to this comment, it sounds like numbers might be calculated differently for the API and the UI. Is that still the case?
Thanks a lot!
GA UI data is picked from the pre-aggregated tables.
Pre-calculated data -- pre-aggregated tables
These are the precalculated data that Google uses to speed up the UI. Google does not specify when this is done but it can be at any point of the time. These are known as pre-aggregated tables
So if you compare the numbers from GA UI to your GA API data, you will always see a discrepancy because the view might be aggregated at some early stages of the day, while your api data is fresh
I would like to add to add a custom extension to my Schedule resource.
In my app, Schedule have visit motives (reasons). I know there's a list of classified appointments / encounter reasons but I would like to use mine.
I have something like this :
{
"resourceType":"Schedule",
"identifier":"logical_id",
"type":"schedule_speciality",
"actor":{
"practioner_id":"identifier",
"practioner_name":"practioner name"
},
"external_id":{
"extension":[
{
"url":"http://api.test.com/fhir/schedule/external_id",
"valueIdentifier":"external_id"
}
]
},
"visit_motives":{
"extension":[
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive1"
},
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive2"
},
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive3"
}
]
},
"practice_id":{
"extension":[
{
"url":"https://api.test.com/fhir/schedule/practice_id",
"valueIdentifier":"practice_id"
}
]
}
}
I'm not sure about this part :
"visit_motives":{
"extension":[
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive1"
},
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive2"
},
{
"url":"https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"valueString":"vist_motive3"
}
]
}
Is it correct to add an extension this way ? There are always multiple visit motives for a specific schedule so I have to list them.
I also have seen this kind of things :
"visit_motives": {
"coding": [
{
"system": "https://api.test.com/fhir/ValueSet/schedule#visit_motives",
"code": "visit_motive1"
}
]
}
Which one is the correct one or am I wrong ?
There are several issues here:
It seems odd to capture a "reason" on a schedule. A schedule says when a particular clinician or clinic or other resource is available. E.g. "Dr. Smith takes appointments Mon/Wed/Fri from 1pm-4pm". So if you were to capture a reason on the resource, it would reflect "Why does Dr. Smith have a schedule?" Typically reasons are captured for an individual Appointment. That's the resource that reserves a particular slot for a planned visit. And Appointment already has an element for reason where you're free to use your own codes or just send text.
You have extensions to convey identifiers, but Schedule already has an element for identifiers. Why would you use extensions instead of the standard element? Note that you can use the "system" and/or "type" components to differentiate different kinds of identifiers.
You're sending "identifier", "type", "name", etc. as simple strings - but they're complex data types, so you need to communicate the child elements
actor is of type Reference - that means you need to point to the Practitioner resource. You can't send the properties in-line. (If the Practitioner only exists in the context of the Schedule, you could use the "contained" approach which would use an internal reference, but containment doesn't seem to make sense in this use-case.
The URL for your extension contains ValueSet, which isn't correct - extensions are all structure definitions. Also, there shouldn't be a # symbol in the URL.
Your syntax for extensions is incorrect. You can't introduce new properties in FHIR. The property name for all extensions is just "extension". You differentiate by the URL. So your syntax should be:
{
"resourceType":"Schedule",
"id":"logical_id",
"extension": [
{
"url":"https://api.test.com/fhir/StructureDefinition/schedule-visit_motive",
"valueString":"vist_motive1"
},
{
"url":"https://api.test.com/fhir/StructureDefinition/schedule-visit_motive",
"valueString":"vist_motive2"
},
{
"url":"https://api.test.com/fhir/StructureDefinition/schedule-visit_motives",
"valueString":"vist_motive3"
}
],
"identifier": [
{
"system": http://api.test.com/fhir/NamingSystem/external_id",
"value": "external_id"
}
{
"system": http://api.test.com/fhir/NamingSystem/practice_id",
"value": "practice_id"
}
]
"type": {
"coding": {
"system": "http://somewhere.org/fhir/CodeSystem/specialties",
"code": "schedule_speciality"
},
"text": "Some text description of specialty"
},
"actor":{
"reference": "http://myserver.org/fhir/Practitioner/12345"
"display": "Dr. smith"
}
}
** UPDATE **
Thanks to Alfred Fuller for pointing out that I need to create a manual index for this query.
Unfortunately, using the JSON API, from a .NET application, there does not appear to be an officially supported way of doing so. In fact, there does not officially appear to be a way to do this at all from an app outside of App Engine, which is strange since the Cloud Datastore API was designed to allow access to the Datastore outside of App Engine.
The closest hack I could find was to POST the index definition using RPC to http://appengine.google.com/api/datastore/index/add. Can someone give me the raw spec for how to do this exactly (i.e. URL parameters, what exactly should the body look like, etc), perhaps using Fiddler to inspect the call made by appcfg.cmd?
** ORIGINAL QUESTION **
According to the docs, "a query can combine equality (EQUAL) filters for different properties, along with one or more inequality filters on a single property".
However, this query fails:
{
"query": {
"kinds": [
{
"name": "CodeProse.Pogo.Tests.TestPerson"
}
],
"filter": {
"compositeFilter": {
"operator": "and",
"filters": [
{
"propertyFilter": {
"operator": "equal",
"property": {
"name": "DepartmentCode"
},
"value": {
"integerValue": "123"
}
}
},
{
"propertyFilter": {
"operator": "greaterThan",
"property": {
"name": "HourlyRate"
},
"value": {
"doubleValue": 50
}
}
},
{
"propertyFilter": {
"operator": "lessThan",
"property": {
"name": "HourlyRate"
},
"value": {
"doubleValue": 100
}
}
}
]
}
}
}
}
with the following response:
{
"error": {
"errors": [
{
"domain": "global",
"reason": "FAILED_PRECONDITION",
"message": "no matching index found.",
"locationType": "header",
"location": "If-Match"
}
],
"code": 412,
"message": "no matching index found."
}
}
The JSON API does not yet support local index generation, but we've documented a process that you can follow to generate the xml definition of the index at https://developers.google.com/datastore/docs/tools/indexconfig#Datastore_Manual_index_configuration
Please give this a shot and let us know if it doesn't work.
This is a temporary solution that we hope to replace with automatic local index generation as soon as we can.
The error "no matching index found." indicates that an index needs to be added for the query to work. See the auto index generation documentation.
In this case you need an index with the properties DepartmentCode and HourlyRate (in that order).
For gcloud-node I fixed it with those 3 links:
https://github.com/GoogleCloudPlatform/gcloud-node/issues/369
https://github.com/GoogleCloudPlatform/gcloud-node/blob/master/system-test/data/index.yaml
and most important link:
https://cloud.google.com/appengine/docs/python/config/indexconfig#Python_About_index_yaml to write your index.yaml file
As explained in the last link, an index is what allows complex queries to run faster by storing the result set of the queries in an index. When you get no matching index found it means that you tried to run a complex query involving order or filter. So to make your query work, you need to create your index on the google datastore indexes by creating a config file manually to define your indexes that represent the query you are trying to run. Here is how you fix:
create an index.yaml file in a folder named for example indexes in your app directory by following the directives for the python conf file: https://cloud.google.com/appengine/docs/python/config/indexconfig#Python_About_index_yaml or get inspiration from the gcloud-node tests in https://github.com/GoogleCloudPlatform/gcloud-node/blob/master/system-test/data/index.yaml
create the indexes from the config file with this command:
gcloud preview datastore create-indexes indexes/index.yaml
see https://cloud.google.com/sdk/gcloud/reference/preview/datastore/create-indexes
wait for the indexes to serve on your developer console in Cloud Datastore/Indexes, the interface should display "serving" once the index is built
once it is serving your query should work
For example for this query:
var q = ds.createQuery('project')
.filter('tags =', category)
.order('-date');
index.yaml looks like:
indexes:
- kind: project
ancestor: no
properties:
- name: tags
- name: date
direction: desc
Try not to order the result. After removing orderby(), it worked for me.