azizalto/vanilla-ml-algorithms

1# src: https://gist.github.com/iamaziz/ea5863beaee090937fd6828e88653f5e2 3 4class LinearRegressionGradient:5    def __init__(self, theta=None):6        self.theta = theta7        self.loss_ = float("inf")8 9    def hypothesis(self, x):10        return self.theta[0] + self.theta[1] * x11 12    def loss(self, X, y):13        m = len(X)14        return sum([(X[i] - y[i]) ** 2 for i in range(m)]) / (2 * m)15 16    def gradientDescent(self, X, y, theta, num_iter=3000, alpha=0.01):17        m = len(X)18 19        for j in range(num_iter):20 21            # predict22            h = list(map(self.hypothesis, X))23 24            # compute slope, aka derivative with current params (theta)25            deri_th0 = sum([(h[i] - y[i]) for i in range(m)]) / m26            deri_th1 = sum([(h[i] - y[i]) * X[i] for i in range(m)]) / m27 28            # update parameters (moving against the gradient 'derivative')29            theta[0] = theta[0] - alpha * deri_th030            theta[1] = theta[1] - alpha * deri_th131 32            # report33            if j % 200 == 0:34                self.loss_ = self.loss(X, y)35                msg = f"loss: {self.loss_}"36                print(msg)37 38 39def app():40    import streamlit as st41 42    def header():43        st.subheader("Linear Regression using Gradient Descent")44        desc = """> Plain Python (vanilla version) i.e. without importing any library"""45        st.markdown(desc)46 47    header()48 49    st1, st2 = st.columns(2)50    with st1:51        code_math()52    with st2:53        interactive_run()54 55    st.markdown(56        f"> source [notebook](https://gist.github.com/iamaziz/ea5863beaee090937fd6828e88653f5e)."57    )58 59 60def code_math():61    import inspect62    import streamlit as st63 64    tex = st.latex65    write = st.write66    mark = st.write67    codify = lambda func: st.code(inspect.getsource(func), language="python")68    cls = LinearRegressionGradient(theta=[0, 0])69 70    write("The class")71    codify(cls.__init__)72 73    write("the Hypothesis")74    tex(r"""h_\theta(x) = \theta_0 + \theta_1x""")75    codify(cls.hypothesis)76    mark('The Loss/Objective/Cost function "_minimize_"')77    tex(r"""J(\theta_0, \theta_1) = \frac{1}{2m}\sum(h_\theta(x^{(i)}) - y^{(i)})^2""")78    codify(cls.loss)79    write("The Gradient Descent algorithm")80    mark("> repeat until converge {")81    tex(82        r"""\theta_0 = \theta_0 - \alpha \frac{1}{m} \sum_{i=1}^{m} (h_\theta(x^{(i)}) - y^{(i)} )"""83    )84    tex(85        r"""\theta_1 = \theta_1 - \alpha \frac{1}{m} \sum_{i=1}^{m} (h_\theta(x^{(i)}) - y^{(i)}) x^{(i)})"""86    )87    mark("> }")88    codify(cls.gradientDescent)89 90 91def interactive_run():92    import streamlit as st93    import numpy as np94 95    mark = st.markdown96    tex = st.latex97 98    def random_data(n=10):99        def sample_linear_regression_dataset(n):100            # src: https://www.gaussianwaves.com/2020/01/generating-simulated-dataset-for-regression-problems-sklearn-make_regression/101            import numpy as np102            from sklearn import datasets103            import matplotlib.pyplot as plt  # for plotting104 105            x, y, coef = datasets.make_regression(106                n_samples=n,  # number of samples107                n_features=1,  # number of features108                n_informative=1,  # number of useful features109                noise=40,  # bias and standard deviation of the guassian noise110                coef=True,  # true coefficient used to generated the data111                random_state=0,112            )  # set for same data points for each run113 114            # Scale feature x (years of experience) to range 0..20115            # x = np.interp(x, (x.min(), x.max()), (0, 20))116 117            # Scale target y (salary) to range 20000..150000118            # y = np.interp(y, (y.min(), y.max()), (20000, 150000))119 120            plt.ion()  # interactive plot on121            plt.plot(x, y, ".", label="training data")122            plt.xlabel("Years of experience")123            plt.ylabel("Salary $")124            plt.title("Experience Vs. Salary")125            # st.pyplot(plt.show())126            # st.write(type(x.tolist()))127            # st.write(x.tolist())128 129            X, y = x.reshape(x.shape[0],), y.reshape(130                y.shape[0],131            )132            return np.around(X, 2), np.around(y, 2)133            # return [a[0] for a in x.tolist()], [a[0] for a in y.tolist()]134            # return [item for sublist in x.tolist() for item in sublist], [135            #     item for sublist in y for item in sublist136            # ]137 138        X_, y_ = sample_linear_regression_dataset(n)139        return X_, y_140        # st.write(type(X_), type(y_))141        # st.write(type(np.round(X, 2).tolist()))142        # st.write(X_)  # , y_)143        # return X, y144 145        # return np.around(X, 2).tolist(), np.around(y, 2).tolist()146 147    X, y = random_data()148    theta = [0, 0]  # initial values149    model = LinearRegressionGradient(theta)150    mark("# Example")151    n = st.slider("Number of samples", min_value=10, max_value=200, step=10)152    if st.button("generate new data and solve"):153        X, y = random_data(n=n)154    mark("_Input_")155    mark(f"_X_ = {X}")156    mark(f"_y_ = {y}")157    model.gradientDescent(X, y, theta)  # run to optimize thetas158    mark("_Solution_")159    tex(f"y = {model.theta[0]:.1f} + {model.theta[1]:.1f} x")  # print solution160    tex(f"loss = {model.loss_}")161 162    mark("> How to run")163    mark(164        """165    ```python166    X, y = random_data()167    theta = [0, 0]  # initial values168    model = LinearRegressionGradient(theta)169    model.gradientDescent(X, y, theta)  # run "i.e. optimize thetas"170    # print solution171    # print(f"y = {model.theta[0]:.1f} + {model.theta[1]:.1f} x")172    # print(f"loss = {model.loss_}")173    ```174    """175    )176    # -- visualize177    import matplotlib.pyplot as plt178 179    fig, ax = plt.subplots()180    ax.scatter(X, y, label="Linear Relation")181    y_pred = theta[0] + theta[1] * np.array(X)182    ax.plot(X, y_pred)183    ax.grid(color="black", linestyle="--", linewidth=0.5, markevery=int)184    ax.legend(loc=2)185    # ax.axis("scaled")186    st.pyplot(fig)187    # st.line_chart(X, y)188