Feat/llm predictor #1
Conversation
915-Muscalagiu-AncaIoana
force-pushed
the
feat/llm_predictor
branch
3 times, most recently
from
c3add2d to
a97f838
Compare
| "article_ids": article_ids, | ||
| } | ||
|
|
||
| def _postprocess_output(self, output): |
There was a problem hiding this comment.
Move the def _postprocess_output(self, output): after def _preprocess_features(self, features): to follow a natural flow of reading.
| - Categorical features: These describe qualitative aspects, like product category, color, and material. | ||
| 3. Your response should only include the probability of purchase for the positive class (e.g., likelihood of being purchased), as a value between 0 and 1. | ||
|
|
||
| ### Product Features: |
There was a problem hiding this comment.
These are all product features? some of them are not of the customers?
There was a problem hiding this comment.
Indeed it's also for the customer.
|
|
||
| scores = [] | ||
| for feature in preprocessed_features: | ||
| langchain_output = self.llm.invoke(feature) |
There was a problem hiding this comment.
Why do we invoke the llm / feature?
There was a problem hiding this comment.
It's indeed not on the feature, but on the feature set of a candidate together with the customer features, which is one data points that needs to be predicted.
| return project, project.get_feature_store() | ||
|
|
||
|
|
||
| def get_secrets_api(): | ||
| connection = hopsworks.connection(host="c.app.hopsworks.ai", | ||
| hostname_verification=False, |
There was a problem hiding this comment.
Have you tried this without adding host="c.app.hopsworks.ai", hostname_verification=False, port=443 ?
You could try to access the get_secrets_api() method directly from the project returned by login. That is usually how it was done throughout the project.
There was a problem hiding this comment.
The problem is that the secrets_api is not linked with the project instance in hopsworks, but it is at the level of the user ( hence you need also tokens with also the user scope checked in order to access it, which is not the default scope of a token ), therefore you have to take it from hopsworks.connection() not from the project. Now, trying this using the default configuration of the connection() produces this error:
File "/Users/ancaioanamuscalagiu/Documents/hands-on-recommender-system/.venv/lib/python3.11/site-packages/hopsworks/client/external.py", line 38, in init
raise exceptions.ExternalClientError("host")
hopsworks.client.exceptions.ExternalClientError: host
which if we look up in their source code can be traced back to:
"""Initializes a client in an external environment such as AWS Sagemaker."""
if not host:
raise exceptions.ExternalClientError("host")
so therefore when it is an external client ( such as my local machine trying to perform this connect ) it expects the host parameter to be set.
| secrets = secrets_api.get_secrets() | ||
| existing_secret_keys = [secret.name for secret in secrets] | ||
| # Create the OPENAI_API_KEY secret if it doesn't exist | ||
| if "OPENAI_API_KEY" not in existing_secret_keys: |
There was a problem hiding this comment.
If this runs on the cloud, the settings will not load the .env file. Thus, it will load an empty value. I would avoid doing this step. Better assert these values and crash the program if they are missing with a clear message.
There was a problem hiding this comment.
A safer option would be to override the settings object at start-up time, based on the Hopsworks secrets. Like this the settings object is the single source of truth.
There was a problem hiding this comment.
To avoid adding these secrets manually, we could create a different scripts that adds them based on the settings file, but I would make it an explicit operation to avoid weird behavior.
| } | ||
|
|
||
| def _postprocess_output(self, output): | ||
| return float(output['text'].split(':')[1].strip()) |
There was a problem hiding this comment.
I would also check if the output is a flow, otherwise do an try expect, returning a minimum score or something to avoid crashing the probram.
you can do that easily with Pydantic classic + LangChain to check that the value is a float within an expected range.
915-Muscalagiu-AncaIoana
force-pushed
the
feat/llm_predictor
branch
2 times, most recently
from
7b5d5d5 to
80c1a99
Compare
915-Muscalagiu-AncaIoana
force-pushed
the
feat/llm_predictor
branch
from
80c1a99 to
093be19
Compare
No description provided.