[Day 186] Prefect cloud, model serving with FastAPI, and SHAP values

 Hello :)
Today is Day 186!


A quick summary of today:
  • using FastAPI for model serving, and running it in Docker
  • starting to think about model monitoring


All code, pictures are on my github repo.


Firstly, yesterday I was having version issues between prefect and prefect-gcp - fixed that with the help of Prefect's community slack, and now I can run one file which runs both flows - upload_to_gcs flow and train_model flow. I also realised that for solo (and mini project) devs Prefect cloud is free, so I uploaded my flows to there just to be a bit safer - not having everything locally only. 

The UI is the same, but it is in the cloud. 

Next, serving the model for inference

I decided to use FastAPI as it seems very easy to create and run a server. Also when I run a fastapi server, there is a nice /docs page where I can quickly and easily test my endpoints. 

It is pretty basic for now, and I just need to improve the prefict_fraud function. I also set up the server to run in docker. Again I had some issues with file location and ports, but I could more easily figure them out (given my experience with Docker from the data eng pipeline project). 
The Dockerfile:

And docker-compose.yml

Given I got this server set up, and also having to serve Prefect flows, I decided to bring in make. 

My makefile has the following options (for now at least):

Maybe if I can figure out how to set up to run Prefect in docker, this will become a bit simpler. But I am yet to figure this out. Nevertheless for now, it is not bad ^^


Now that I had model serving (with FastAPI) set up, I started to think about monitoring.

But what can I monitor? The dataset is not only small, but no new data will come in. I can of course set aside some rows of data, or even generate fake data maybe. 

I can monitor:

  • performance metrics (recall, accuracy, roc-auc, false positive rate, false negative rate, etc)
  • fraud/not fraud distribution
  • look for anomalies
  • optimal threshold
  • feature importance

One interesting and new (for me) way to look at feature importance is to look at SHAP values. SHAP shows feature importance based on the contribution of each feature to the model's predictions, averaged over all possible feature combinations, ensuring a fair and consistent representation of their impact. 

After some issues with generating visualisations, I got some charts

Fraud SHAP values:

Not Fraud SHAP values:

There are some other png files that I managed to create (following this video). All are on my repo. 


On another note ~ I realised I might have made a mistake when preparing the dataset I used for the model. What I did was, I had all these variables: NumberOfSuppliments,AgeOfVehicle,AgeOfPolicyHolder,Month,Deductible,MonthClaimed,Make,AddressChange_Claim,PastNumberOfClaims,VehiclePrice,VehicleCategory,Fault,FraudFound_P

And I just got dummies for all. However, I did this without thinking whether ordinal encoding might be better. After some review, doing dummies for all seems fine, but for other cases I need to keep this in mind. Maybe it is something to explore - doing ordinal encoding for some of the vars for which it might be interesting to see the effect on model performance (like vehicle price maybe). Nevertheless, for now doing OHE(dummies) for all seems fine, and the model performance is satisfactory for now.


That is all for today!

See you tomorrow :)

Popular posts from this blog

[Day 198] Transactions Data Streaming Pipeline Porject [v1 completed]

Image
 Hello :) Today is Day 198! A quick summary of today: data streaming pipeline project [v1 done] Here is a link to the project's repo. Well ... I did not know I can do it in a day (~14 hours) after yesterday's issues but here we are. Turns out in order to insert the full (~70 variables with nested/list structure), I need the proper pyspark schema. And yesterday I did not have that and that is why when I was reading data in the kafka producer I was getting NULL in the columns - my schema was wrong. Well today I not only fixed the schema for the 4 variables I had yesterday, but included *all* the variables that come from the Stripe API ~ 70 (for completeness).  When I run docker-compose, the data streams and is input into the postgres db (and is still running). Unfortunately, the free Stripe API for creating realistic transactions has a limit of 25, so every 3 seconds, 25 new transactions are sent to the db. It has been running half the day (since I got that set up) and as I am wr
Read more

[Day 107] Transforming natural language to charts

Image
 Hello :) Today is Day 107! A quick summary of today: created text2chart - transforming natural language to charts [ github ] [ webapp ] I am super sick, but in the around 3 hours of my fever not being high, I managed to study a bit. I was again thinking about creating some nlp project but was stuck on the idea of how to host a model. After looking around I found this article where they create a similar app called chat2vis  and they use codellama. I know codellama is an open source model that I can get from huggingface and I wondered, how did they host it. Turns out - they ask the user to input a huggingface api key, which then allows for querying and generating content. (AWESOME! this is applicable to some of the other PDF chat apps that I have, but I will look to make an app of them later).  My fever eased a bit, and I sat on my chair to do some work ^^ Here is a breakdown of the app. I used streamlit. Also, 2 default datasets during dev: Financial Statements of Major Companies(2009-
Read more

[Day 54] I became a backprop ninja! (woohoo)

Image
 Hello :) Today is Day 54! A quick summary of today: Read through cs231n's neural network lectures [1] [2] [3] Spent 8 hours going over Andrej Karpathy's backprop ninja tutorial After yesterday's confidence boost in doing backprop, I came in today with the intention of becoming of, as Andrej Karpathy calls it, a backprop ninja.  The cs231n NN lectures were the usual about how NN became to be, param initialization, activation functions, loss functions, regularization and optimizers. Nothing out of the ordinary, but still was good to reassure myself of that knowledge. But that was definitely not the main even of today. As I mentioned, I wanted to become (or at least attempt to become) a backprop ninja today. I have watched that particular tutorial twice, the 1st time I had no idea what is going on, the 2nd time I could follow, and now it was the 3rd time - and now, I wanted to participate and actually see the why and the how of the things we have to do, when we do manual back
Read more