import numpy as np
from scipy.sparse.linalg import lsqr
from sklearn.datasets import load_iris, load_breast_cancer
from sklearn.linear_model import RidgeClassifier as skRidgeClassifier

class RidgeClassifier:
    def __init__(self, alpha=1.0):
        self.alpha = alpha

    def _encode(self, y):
        classes = np.unique(y)
        y_train = np.full((y.shape[0], len(classes)), -1)
        for i, c in enumerate(classes):
            y_train[y == c, i] = 1
        if len(classes) == 2:
            y_train = y_train[:, 1].reshape(-1, 1)
        return classes, y_train

    def _solve_lsqr(self, X, y):
        coefs = np.zeros((y.shape[1], X.shape[1]))
        for i in range(y.shape[1]):
            cur_y = y[:, i]
            info = lsqr(X, cur_y, np.sqrt(self.alpha))
            coefs[i] = info[0]
        return coefs

    def fit(self, X, y):
        self.classes_, y_train = self._encode(y)
        X_mean = np.mean(X, axis=0)
        y_mean = np.mean(y_train, axis=0)
        X_train = X - X_mean
        y_train = y_train - y_mean
        self.coef_ = self._solve_lsqr(X_train, y_train)
        self.intercept_ = y_mean - np.dot(X_mean, self.coef_.T)
        return self

    def decision_function(self, X):
        scores = np.dot(X, self.coef_.T) + self.intercept_
        if scores.shape[1] == 1:
            return scores.ravel()
        else:
            return scores

    def predict(self, X):
        scores = self.decision_function(X)
        if len(scores.shape) == 1:
            indices = (scores > 0).astype(int)
        else:
            indices = np.argmax(scores, axis=1)
        return self.classes_[indices]

# binary classification
for alpha in [0.1, 1, 10]:
    X, y = load_breast_cancer(return_X_y = True)
    clf1 = RidgeClassifier(alpha=alpha).fit(X, y)
    clf2 = skRidgeClassifier(alpha=alpha, solver='lsqr',
                             # keep consisent with scipy default
                             tol=1e-8).fit(X, y)
    assert clf1.coef_.shape == (1, X.shape[1])
    assert np.allclose(clf1.coef_, clf2.coef_)
    assert np.allclose(clf1.intercept_, clf2.intercept_)
    prob1 = clf1.decision_function(X)
    prob2 = clf2.decision_function(X)
    pred1 = clf1.predict(X)
    pred2 = clf2.predict(X)
    assert np.allclose(prob1, prob2)
    assert np.array_equal(pred1, pred2)

# multiclass classification
for alpha in [0.1, 1, 10]:
    X, y = load_iris(return_X_y = True)
    clf1 = RidgeClassifier(alpha=alpha).fit(X, y)
    clf2 = skRidgeClassifier(alpha=alpha, solver='lsqr',
                             # keep consisent with scipy default
                             tol=1e-8).fit(X, y)
    assert clf1.coef_.shape == (len(np.unique(y)), X.shape[1])
    assert np.allclose(clf1.coef_, clf2.coef_)
    assert np.allclose(clf1.intercept_, clf2.intercept_)
    prob1 = clf1.decision_function(X)
    prob2 = clf2.decision_function(X)
    pred1 = clf1.predict(X)
    pred2 = clf2.predict(X)
    assert np.allclose(prob1, prob2)
    assert np.array_equal(pred1, pred2)