Creating a logistic regression classifier involves pretty much the same steps as setting up k-NN:

In [14]: lr = cv2.ml.LogisticRegression_create()

We then have to specify the desired training method. Here, we can choose cv2.ml.LogisticRegression_BATCH or cv2.ml.LogisticRegression_MINI_BATCH. For now, all we need to know is that we want to update the model after every data point, which can be achieved with the following code:

In [15]: lr.setTrainMethod(cv2.ml.LogisticRegression_MINI_BATCH)
...      lr.setMiniBatchSize(1)

We also want to specify the number of iterations the algorithm should run before it terminates:

In [16]: lr.setIterations(100)

We can then call the train method of the object (in the exact same way as we did earlier), which will return True upon success:

In [17]: lr.train(X_train, cv2.ml.ROW_SAMPLE, y_train)
Out[17]: True

As we just saw, the goal of the training phase is to find a set of weights that best transform the feature values into an output label. A single data point is given by its four feature values (f₀, f₁, f₂, and f₃). Since we have four features, we should also get four weights, so that x = w₀ f₀ + w₁ f₁ + w₂ f₂ + w₃ f₃, and ŷ=σ(x). However, as discussed previously, the algorithm adds an extra weight that acts as an offset or bias, so that x = w₀ f₀ + w₁ f₁ + w₂ f₂ + w₃ f₃ + w₄. We can retrieve these weights as follows:

In [18]: lr.get_learnt_thetas()
Out[18]: array([[-0.04090132, -0.01910266, -0.16340332, 0.28743777, 0.11909772]], dtype=float32)

This means that the input to the logistic function is x = -0.0409 f₀ - 0.0191 f₁ - 0.163 f₂ + 0.287 f₃ + 0.119. Then, when we feed in a new data point (f₀, f₁, f₂, f₃) that belongs to class 1, the output ŷ=σ(x) should be close to 1. But how well does that actually work?