Training killed automatically for older messages, why and /samples api is helping

[onkar-infoobjects]

Hello,

Message "Can you name a good law firm in Chennai?" was trained like showing below. I have picked up the entities from expressions.json file the downloaded data.

{
"text" : "Can you name a good law firm in Chennai?",
"entities" : [
{
"entity" : "action",
"value" : "business"
},
{
"entity" : "intent",
"value" : "fetchItems"
},
{
"entity" : "bizCategory",
"value" : "law firm",
"start" : 21,
"end" : 29
},
{
"entity" : "wit$location",
"value" : "Chennai",
"start" : 33,
"end" : 40
}
]
}

We trained this message few days back but when i looked today, i did not get the entities which were trained.

And this is happening with many messages. Its a critical point. We are loosing the efforts we put on.

And i am trying to train those messages with /samples api and that api is sending correct response but still wit response for the message is same.

What should be done in this case?

[blandinw]

@onkar-infoobjects Apologies for the late reply. There's probably a logical explanation to this, no need to panic.
Please follow the issue template so that we can better help you. What is your app id?

[onkar-infoobjects]

Hey @blandinw ,

Sorry i missed the notification.
The App id is: 588b1027-6be1-4ab0-818e-649c66abb958

Would you be able to let me know if i am doing something wrong with the training?

[onkar-infoobjects]

Hey @blandinw,

Any luck?

[gulbinas]

App ID: 5a11f551-d02e-4fdf-a379-b35db88e2d74

Appears in aforementioned app we have same issues.

Looks like there is misconception how to use /samples API endpoint.

To my understanding this endpoint should emulate consoles "Verify" button, am I right?

[blandinw]

The POST /samples API queues up requests to be processed asynchronously when the system is under pressure. The system experienced slowness lately, hence POST /samples returning a 200 and app not being trained right away. We're implementing 2 things: a) massive speed ups of queue processing and b) training status API for you to know when the training actually will happen or happened.

[gulbinas]

@blandinw in that case, this is still useful: wit-ai/pywit#101

[stopachka]

Both issues, and the new training API tracked on #876