azure / aml-deploy Goto Github PK

@marvinbuss the syntax {"IS_SENT_START": False, "OP": "*"} works only for words that come before the word in the pattern right? what about the sentences where we can't make any assumptions about how the sentence will end but still want to match as soon as the sentence starts with a certain set of words. One example: "I thought I share our newest features with you" Now the Words we match is "I thought" and "our". Now do I need to add something to make sure it matches even though there are words coming after the words in pattern3?

` def _you_centered21(
self,
rule_id="R0061",
description="Put your recipient in the center of your sentence. Simply replace the I focused sentence with a You focused one.That's how you catch their attention.",
info="Try to be very precise on what you want and use the present form if possible.",
example_pos="You can find a guide on how you can benefit from the most recent product update attached.",
example_neg="I thought I share our newest features with you"
) -> None:
""" Check for negative and positive tone.

    rule_id (str): The ID of the rule.
    description (str): The description of the rule.
    info (str): Additional info about the rule.
    example_pos (str): Positive example for this rule.
    example_neg (str): Negative example for this rule.
    """
    # Define matcher
    matcher = Matcher(self.nlp.vocab)

    # Define patterns to match
    pattern1 = [
        {"LOWER": "I"}
    ]

    pattern2 = [
        {"LOWER": "thought"}
    ]

    pattern3 = [
        {"LOWER": "our"} ,{"IS_SENT_START": False, "OP": "*"}
    ]


    # Add the patterns to the matcher and apply the matcher to the doc
    matcher.add("positivetone_pattern", [pattern1, pattern2, pattern3])
    matches = matcher(self.doc)

    # Merge the matched intervals
    intervals = [(start, end) for match_id, start, end in matches]
    merged = merge_intervals(intervals)

    # Add recommendations
    if len(merged) > 0:
        self._add_recommendation(
            rule_id=rule_id,
            description=description,
            info=info,
            example_pos=example_pos,
            example_neg=example_neg,
            token_intervals=merged
        )`

Hi,

I have a registered model that I want to deploy, but when trying to load it with:

in score.py:

import tensorflow.keras.models import load_model
model_path = os.path.join(os.getenv("AZUREML_MODEL_DIR"), 
                            "outputs", "best_model")
model = load_model(model_path)

I get the error that saved_model.pb is not found in the folder. If you look in registered model/Artifacts/ the file saved_model.pb exists

raise IOError("SavedModel file does not exist at: %s/{%s|%s}" %
OSError: SavedModel file does not exist at: /var/azureml-app/azureml-models/p8_model/27/outputs/best_model/{saved_model.pbtxt|saved_model.pb}

What can be the issue here?

I just realized that azureml copies the model folder hierarchy inconsistently from one deploy to another. How to know in what folder the final model will be? (For one deploy it was outputs/best_model, for another it was without outputs folder)

Regards,
Voka

Hi @marvinbuss , I noticed there are functions with single words ergo only one pattern. Will the model detect sentences with the words or does the function only work when there are sentences with one word. I inserted a function for you to check.

` def _positive_word08(
self,
rule_id="R0079",
description="The word ""But"" implies a conflict. It will subconsciously affect your recipient. To make sure you keep it positive use a stronger, more confident word if possible. ",
info="Try to be very precise on what you want and use the present form if possible.",
example_pos="And",
example_neg="But"
) -> None:
""" Check for negative and positive tone.

    rule_id (str): The ID of the rule.
    description (str): The description of the rule.
    info (str): Additional info about the rule.
    example_pos (str): Positive example for this rule.
    example_neg (str): Negative example for this rule.
    """
    # Define matcher
    matcher = Matcher(self.nlp.vocab)

    # Define patterns to match
    pattern1 = [
        {"LOWER": "But"}
    ]


    # Add the patterns to the matcher and apply the matcher to the doc
    matcher.add("positivetone_pattern", [pattern1])
    matches = matcher(self.doc)

    # Merge the matched intervals
    intervals = [(start, end) for match_id, start, end in matches]
    merged = merge_intervals(intervals)

    # Add recommendations
    if len(merged) > 0:
        self._add_recommendation(
            rule_id=rule_id,
            description=description,
            info=info,
            example_pos=example_pos,
            example_neg=example_neg,
            token_intervals=merged
        )` @

az ml model deploy [--namespace] with the required information but the namespace doesn't get reflected in the url of the deployed endpoint in aks

Hi ,
AML supports "controlled rollout" to AKS - https://docs.microsoft.com/en-us/azure/machine-learning/how-to-deploy-azure-kubernetes-service#deploy-models-to-aks-using-controlled-rollout-preview.

In summary, this feature allows, multiple versions of model to be deployed on same AKS cluster. Each version of the model serves some percentage of traffic for the same endpoint.

Is there a plan to add support for this feature?

The user must be able to pass a list for the following input values:

• model_name
• model_version
  This is required to deploy multiple models to one container.

the docker image creation option is available but once image is created and pushed into the container registry, if you try to deploy the image, it does not work, reason is image is just the model package image and does not contain the web server code which comes from mcr.microsoft.com/azureml/aci-init-container
and mcr.microsoft.com/azureml-aci-frontend container.
these containers are deployed along with model package by Azure ML when ACI deployment is selected. It also requires a lot of environment params written by Azure ml

now user can not use currently available actions to deploy this image, because those all details are hidden.

can we provide a mechanism for user to deploy these images?

Hi,

I am encountering the below error when I am trying to deploy a model in ACI.

Traceback (most recent call last):
  File "/code/main.py", line 318, in <module>
    main()
  File "/code/main.py", line 129, in main
    raise AMLConfigurationException(f"Could not load model with provided details: ***exception***")
utils.AMLConfigurationException: Could not load model with provided details: WebserviceException:
	Message: ModelNotFound: Model with name porto_seguro_safe_driver_model, version 14 not found in provided workspace
	InnerException None
	ErrorResponse 
***
    "error": ***
        "message": "ModelNotFound: Model with name porto_seguro_safe_driver_model, version 14 not found in provided workspace"
    ***
***

Here is my GitHub Action yaml.

      - name: Deploy Model to ACI
        id: deploy_aci
        uses: Azure/aml-deploy@v1
        with:
          azure_credentials: ${{ secrets.AZURE_CREDENTIALS }}
          model_name:  "${{ env.MODEL_NAME }}"
          model_version: "${{ env.MODEL_VERSION }}"
          parameters_file: "deploy_aci.json"

I verified that I have these models in my AML.

ka@Azure:~$ az ml model list -o table
Name                            Version    Framework    CreatedTime
------------------------------  ---------  -----------  -------------------
porto_seguro_safe_driver_model  14         Custom       2020-11-20T15:16:19
porto_seguro_safe_driver_model  13         Custom       2020-11-19T17:25:36
porto_seguro_safe_driver_model  12         Custom       2020-11-19T16:39:02
porto_seguro_safe_driver_model  11         Custom       2020-11-19T16:15:37
porto_seguro_safe_driver_model  10         Custom       2020-09-18T17:25:55
porto_seguro_safe_driver_model  9          Custom       2020-09-04T15:55:04
porto_seguro_safe_driver_model  8          Custom       2020-09-03T22:40:08
porto_seguro_safe_driver_model  7          Custom       2020-09-03T21:47:30
porto_seguro_safe_driver_model  6          Custom       2020-09-03T16:11:49
porto_seguro_safe_driver_model  5          Custom       2020-09-03T14:08:11
porto_seguro_safe_driver_model  4          Custom       2020-08-28T21:22:35
porto_seguro_safe_driver_model  3          Custom       2020-08-28T20:38:17
driver_model.pkl                2          Custom       2020-08-28T20:17:31
porto_seguro_safe_driver_model  2          Custom       2020-08-25T02:49:24
porto_seguro_safe_driver_model  1          Custom       2020-08-25T02:21:00
driver_model.pkl                1          Custom       2020-08-19T18:04:23

Any suggestion on how I should go about this error?

• Deploy mode from release

• Create a docker image of the model

Please add another test for deployments to Azure Functions.

when i tried running this code i got this error, kindly assist on it
import json

input_payload = json.dumps({
'data': [[34.927778, 0.24, 7.3899, 83, 16.1000, 1016.51, 1]],
'method': 'predict' # If you have a classification model, you can get probabilities by changing this to 'predict_proba'.
})

output = service.run(input_payload)

print(output)

error :

Received bad response from service. More information can be found by calling .get_logs() on the webservice object.
Response Code: 502
Headers: {'Connection': 'keep-alive', 'Content-Length': '38', 'Content-Type': 'text/html; charset=utf-8', 'Date': 'Tue, 26 Apr 2022 05:35:50 GMT', 'Server': 'nginx/1.10.3 (Ubuntu)', 'X-Ms-Request-Id': 'd84f9266-f41c-4587-a3e9-56b482cccc80', 'X-Ms-Run-Function-Failed': 'True'}
Content: b"'str' object has no attribute 'tolist'"

WebserviceException Traceback (most recent call last)
in
7 })
8
----> 9 output = service.run(input_payload)
10
11 print(output)

/anaconda/envs/azureml_py38/lib/python3.8/site-packages/azureml/core/webservice/aci.py in run(self, input_data)
398 return resp.json()
399 else:
--> 400 raise WebserviceException('Received bad response from service. More information can be found by calling '
401 '.get_logs() on the webservice object.\n'
402 'Response Code: {}\n'

WebserviceException: WebserviceException:
Message: Received bad response from service. More information can be found by calling .get_logs() on the webservice object.
Response Code: 502
Headers: {'Connection': 'keep-alive', 'Content-Length': '38', 'Content-Type': 'text/html; charset=utf-8', 'Date': 'Tue, 26 Apr 2022 05:35:50 GMT', 'Server': 'nginx/1.10.3 (Ubuntu)', 'X-Ms-Request-Id': 'd84f9266-f41c-4587-a3e9-56b482cccc80', 'X-Ms-Run-Function-Failed': 'True'}
Content: b"'str' object has no attribute 'tolist'"
InnerException None
ErrorResponse
{
"error": {
"message": "Received bad response from service. More information can be found by calling .get_logs() on the webservice object.\nResponse Code: 502\nHeaders: {'Connection': 'keep-alive', 'Content-Length': '38', 'Content-Type': 'text/html; charset=utf-8', 'Date': 'Tue, 26 Apr 2022 05:35:50 GMT', 'Server': 'nginx/1.10.3 (Ubuntu)', 'X-Ms-Request-Id': 'd84f9266-f41c-4587-a3e9-56b482cccc80', 'X-Ms-Run-Function-Failed': 'True'}\nContent: b"'str' object has no attribute 'tolist'""
}
}

is user gives service name as which does not confirm to Azure restrictions for naming. Should we make the whole name as lowercase?

##[error]Model deployment failed with exception: WebserviceException:
Message: Error, provided service name is invalid. It must only consist of lowercase letters, numbers, or dashes, start with a
letter, end with a letter or number, and be between 3 and 32 characters long.
InnerException None
ErrorResponse

"error": ***
    "message": "Error, provided service name is invalid. It must only consist of lowercase letters, numbers, or dashes, start with a letter, end with a letter or number, and be between 3 and 32 characters long."
***

input given : "testACIService"

• Enable deployment to Azure function

This is in relation to #6 & #8
With the creation of an image into the AML Workspace related ACR, multiple Repos per image per aml-deploy call will be created.
It seems, that calling Model.package(...) is the core reason. (create_image related)
Maybe you could state this somewhere in the documentation or implement a workaround if there is low effort.
Here an example of my ACR repo list:

while deploying a model to AKS using aml-deploy action if the compute target cluster is not found in the workspace it then switches to ACI deployment for the model.

Hi Everyone,

Please I have my Databricks instance in a VNET. I’m trying to deploy my Machine learning model using the Azure ML workspace on Azure Container Instance (ACI).

I’m able to create an ML workspace. I get an SSLERROR when I try to register the Model using Model.register().

Using this code -

`from azureml.core import Workspace
from azureml.core.model import Model
import azureml.core
from azureml.core.workspace import Workspace
from azureml.core.model import Model

from azureml.core import Workspace
ws = Workspace.create(name='myworkspace',
subscription_id='mysub_id',
resource_group='myresourcegroup',
location='eastus'
)

model_reg = Model.register(model_path = “./model_dir”,
model_name = "ModelX",
workspace = ws)
`

Find below the error when I try to deploy my model.

SSL Error:
SSLError: HTTPSConnectionPool(host='eastus.experiments.azureml.net', port=443): Max retries exceeded with url: /discovery (Caused by SSLError(SSLError("bad handshake: SysCallError(104, 'ECONNRESET')")))
Please note only the Azure Databricks is in a VNET on Azure. How do I resolve it and deploy my model as a webservice on ACI.

Thank you.

this links to this issue
https://github.com/github/ms-ecosystem/issues/85

Model profiling would be a nice feature to add. Alternatively, we could create a new Action for this feature to simplify the Action.