Logging Model Signature¶

Another important part of logging a model is to log it's signature. ¿What is a model signature? A model signature is a description of the input and output of the model. This is important because it allows us to know what kind of data the model expects and what kind of data it will return.

Imagine the input of the model is a 2D array of shape (n_samples, n_features) and the output is a 1D array of shape (n_samples). The signature of this model would be:

In [1]:

import numpy as np


inputs = np.random.randn(1, 100)  # 1 sample of 100 features
outputs = np.random.randn(1, )  # 1 sample

import numpy as np inputs = np.random.randn(1, 100) # 1 sample of 100 features outputs = np.random.randn(1, ) # 1 sample

We generate the signature of the model using the infer_signature function from the mlflow.models.signature module. This function receives two parameters, the input and the output of the model. The input and output of the model are represented as a pandas DataFrame. The infer_signature function will return a string with the signature of the model.

In [2]:

from mlflow.models import infer_signature


signature = infer_signature(model_input=inputs, model_output=outputs)

from mlflow.models import infer_signature signature = infer_signature(model_input=inputs, model_output=outputs)

We can log the model signature along with the model using the log_model function.

In [3]:

from sklearn.linear_model import LinearRegression
import mlflow


model = LinearRegression()


with mlflow.start_run():

    # signature is logged along with the model
    mlflow.sklearn.log_model(model, "model", signature=signature)

from sklearn.linear_model import LinearRegression import mlflow model = LinearRegression() with mlflow.start_run(): # signature is logged along with the model mlflow.sklearn.log_model(model, "model", signature=signature)