m2cgen dApp

An Machine Learning example using Cartesi Rollups

This example shows a simple way of leveraging some of the most widely used Machine Learning libraries available in Python.

The dApp generates a logistic regression model using scikit-learn, NumPy and pandas, and then uses m2cgen (Model to Code Generator) to transpile that model into native Python code with no dependencies. This approach is inspired by Davis David's Machine Learning tutorial, and is useful for a Cartesi DApp because it removes the need of porting all those Machine Learning libraries to the Cartesi Machine's RISC-V architecture, making the development process easier and the final back-end code simpler to execute.

The practical goal of this application is to predict a classification based on the Titanic dataset, which shows characteristics of people onboard the Titanic and whether that person survived the disaster or not. As such, users can submit inputs describing a person's features to find out if that person is likely to have survived.

The model currently takes into account only three characteristics of a person to predict their survival, even though other attributes are available in the dataset:

1. Age
2. Sex, which can be male or female
3. Embarked, which corresponds to the port of embarkation and can be C (Cherbourg), Q (Queenstown), or S (Southampton)

As such, inputs to the DApp should be given as a JSON string such as the following:

{ "Age": 37, "Sex": "male", "Embarked": "S" }

The predicted classification result will be given as 0 (did not survive) or 1 (did survive).

Environment Setup - Requirements

To use this dApp, you need to set up your environment for Cartesi Rollups development with Cartesi CLI. Below is a link to guide you through the environment setup:

1. Installation guide

Building and running the application

To build this application, you should have Cartesi CLI installed. With that, simply clone this repository and in the root folder open a terminal window and run:

cartesi build

Give it some minutes to build the docker containers. After that, run :

cartesi run

After running the containers, your terminal window will be similar to the below:

Attaching to f10a55b8-prompt-1, f10a55b8-validator-1
f10a55b8-prompt-1     | Anvil running at http://localhost:8545
f10a55b8-prompt-1     | GraphQL running at http://localhost:8080/graphql
f10a55b8-prompt-1     | Inspect running at http://localhost:8080/inspect/
f10a55b8-prompt-1     | Explorer running at http://localhost:8080/explorer/
f10a55b8-prompt-1     | Press Ctrl+C to stop the node

Interacting with the application

We can use the cartesi send generic command application to interact with the DApp.

First, go to a separate terminal window and send an input. Your terminal window will be similar to below:

? Chain (Use arrow keys)
❯ Foundry
? Chain Foundry
? RPC URL http://127.0.0.1:8545
? Wallet Mnemonic
? Mnemonic test test test test test test test test test test test junk
? Account 0x3C44CdDdB6a900fa2b585dd299e03d12FA4293BC 10000 ETH
? DApp address 0x70ac08179605AF2D9e75782b8DEcDD3c22aA4D0C
? Input String encoding
? Input (as string) { "Age": 37, "Sex": "male", "Embarked": "S" }
✔ Input sent: 0x02b7506ef56d5560fcc34f09af7de299275a54e42a157b26d1911f7b4b83e09c

And your running backend window will be similar to below:

f10a55b8-validator-1  | [INFO  rollup_http_server::http_service] Received new request of type ADVANCE
f10a55b8-validator-1  | [INFO  actix_web::middleware::logger] 127.0.0.1 "POST /finish HTTP/1.1" 200 290 "-" "python-requests/2.31.0" 0.001344
f10a55b8-validator-1  | INFO:__main__:Received finish status 200
f10a55b8-validator-1  | INFO:__main__:Received advance request data {'metadata': {'msg_sender': '0x3c44cdddb6a900fa2b585dd299e03d12fa4293bc', 'epoch_index': 0, 'input_index': 0, 'block_number': 12, 'timestamp': 1704202333}, 'payload': '0x7b2022416765223a2033372c2022536578223a20226d616c65222c2022456d6261726b6564223a20225322207d'}
f10a55b8-validator-1  | INFO:__main__:Received input: '{ "Age": 37, "Sex": "male", "Embarked": "S" }'
f10a55b8-validator-1  | INFO:__main__:Data={ "Age": 37, "Sex": "male", "Embarked": "S" }, Predicted: 0
f10a55b8-validator-1  | INFO:__main__:Adding notice with payload: 0
f10a55b8-validator-1  | [INFO  actix_web::middleware::logger] 127.0.0.1 "POST /notice HTTP/1.1" 201 11 "-" "python-requests/2.31.0" 0.000896
f10a55b8-validator-1  | INFO:__main__:Received notice status 201 body b'{"index":0}'
f10a55b8-validator-1  | INFO:__main__:Sending finish

The payload of the notice corresponds to whether the person survived ("1") or not ("0"). In this case, it should be "0".

Changing the application

This dApp was implemented in a rather generic way and, as such, it is possible to easily change the target dataset as well as the predictor algorithm.

To change those, open the file m2cgen/model/build_model.py and change the following variables defined at the beginning of the script:

• model: defines the scikit-learn predictor algorithm to use. While it currently uses sklearn.linear_model.LogisticRegression, many other possibilities are available, from several types of linear regressions to solutions such as support vector machines (SVMs).
• train_csv: a URL or file path to a CSV file containing the dataset. It should contain a first row with the feature names, followed by the data.
• include: an optional list indicating a subset of the dataset's features to be used in the prediction model.
• dependent_var: the feature to be predicted, such as the entry's classification