A repo for a small (py)opencl project

I will be implementing stochastic gradient descent for logistic regression using OpenCL.

---

OLS Gradient descent algorithm

• OLS = ordinary least squares
• The objective of this algorithm is to estimate model parameters by minimizing the sum of squares between the model estimates and training examples
• Cost function:

$$ J(x,y,\beta) = \frac{1}{2} \sum_{i=1}{N} (h(x_{i}, \beta) - y_{i})^{2} $$

Where $h(.)$ is the hypothesis function (a.k.a. model).

• Since this is a convex function, this is minimized for the parameters $\beta$ when:

$$ \frac{\partial J}{\partial \beta} = 0 \newline \sum_{i=1}^{N} (h(x_{i}, \beta)) * \frac{\partial h(x_{i}, \beta)}{\partial beta} = 0 $$