Scikit-learn

Creating a simple linear regression model using scikit-learn and using dstack to pull and push the model.

We will use the sklearn.datasets package to use the diabetes dataset to make and deploy a simple Linear Regression Model using dstack, then pull it to make a linear regression plot using matplotlib

1. Importing Libraries and dstack

We will first import scikit-learn, numpy and matplotlib for plotting and of course the push and pull methods from dstack

import matplotlib.pyplot as plt
import numpy as np
from sklearn import datasets, linear_model
from sklearn.metrics import mean_squared_error, r2_score
from sklearn.linear_model import LinearRegression
import sklearn
​
import dstack as ds

2. Loading and Splitting Dataset

Let's load our diabetes dataset now from scikit-learn, and split it.

# Loading diabetes databse
diabetes_X, diabetes_y = datasets.load_diabetes(return_X_y=True)
​
# Use only one feature
diabetes_X = diabetes_X[:, np.newaxis, 2]
​
# Split the data into training/testing sets
diabetes_X_train = diabetes_X[:-20]
diabetes_X_test = diabetes_X[-20:]
​
# Split the targets into training/testing sets
diabetes_y_train = diabetes_y[:-20]
diabetes_y_test = diabetes_y[-20:]

3. Fitting the Model

Finally we fit the model with the LinearRegression() object in scikit-learn

# Create linear regression object
regr = LinearRegression()
​
# Fitting the linear model
regr.fit(diabetes_X_train, diabetes_y_train)

4. Pushing to dstack

Now that our model is fit and ready, we push it to dstack as a stack using the push_frame()method

# Push the frame
ds.push("simpleLinearReg", regr, "My first linear model")

5. Pulling from dstack

Now that you have it pushed on a Stack, you can share it with anyone so they can pull the model and use it, or you can pull it and re-use it anytime you like as well! Let's see how we can pull the model from the Stack and use it to display a plot.

# Pull from the Stack
my_model = ds.pull("simpleLinearReg")

That's it! You have pulled your model. It's that easy.

6. Using the Model to Make Predictions or Plots

Now we can use the model to make predictions and use Matplotlib for plotting.

# Make predictions using the testing set
diabetes_y_pred = my_model.predict(diabetes_X_test)
​
# Plot outputs
plt.scatter(diabetes_X_test, diabetes_y_test, color='black')
plt.plot(diabetes_X_test, diabetes_y_pred, color='blue', linewidth=3)
​
plt.xticks(())
plt.yticks(())
​
plt.show()

You should see the following output.

Linear Regression Plot Output

You can also push this plot onto dstack and create a report with the model, plot as well as the dataset! Read the tutorial on the plotting libraries to try this out.

Common Problems and Debugging

AttributeError

This error probably means you haven't called .fit() method on the Linear Regression model.

AttributeError: 'LinearRegression' object has no attribute 'coef_'