import numpy as np
from math import ceil, floor
from sklearn.datasets import load_iris
from sklearn.model_selection import StratifiedShuffleSplit as skStratifiedShuffleSplit

class StratifiedShuffleSplit():
    def __init__(self, n_splits=10,
                 train_size=0.9, test_size=0.1, random_state=0):
        self.n_splits = n_splits
        self.train_size = train_size
        self.test_size = test_size
        self.random_state = random_state

    def _approximate_mode(self, class_counts, n_draws, rng):
        continuous = n_draws * class_counts / class_counts.sum()
        floored = np.floor(continuous)
        need_to_add = int(n_draws - floored.sum())
        if need_to_add > 0:
            remainder = continuous - floored
            values = np.sort(np.unique(remainder))[::-1]
            for value in values:
                inds = np.where(remainder == value)[0]
                add_now = min(len(inds), need_to_add)
                inds = rng.choice(inds, size=add_now, replace=False)
                floored[inds] += 1
                need_to_add -= add_now
                if need_to_add == 0:
                    break
        return floored.astype(int)

    def split(self, X, y):
        n_train = np.floor(self.train_size * X.shape[0])
        n_test = np.ceil(self.test_size * X.shape[0])
        classes, y_indices = np.unique(y, return_inverse=True)
        class_counts = np.bincount(y_indices)
        # quick sort is not stable
        class_indices = np.split(np.argsort(y_indices, kind='mergesort'),
                                 np.cumsum(class_counts)[:-1])
        rng = np.random.RandomState(self.random_state)
        for _ in range(self.n_splits):
            train, test = [], []
            n_i = self._approximate_mode(class_counts, n_train, rng)
            t_i = self._approximate_mode(class_counts - n_i, n_test, rng)
            for i in range(classes.shape[0]):
                permutation = rng.permutation(class_counts[i])
                train.extend(class_indices[i][permutation][:n_i[i]])
                test.extend(class_indices[i][permutation][n_i[i]:n_i[i] + t_i[i]])
            train = rng.permutation(train)
            test = rng.permutation(test)
            yield train, test

X, y = load_iris(return_X_y=True)
cv1 = StratifiedShuffleSplit(n_splits=5, random_state=0)
cv2 = skStratifiedShuffleSplit(n_splits=5, random_state=0)
for (train1, test1), (train2, test2) in zip(cv1.split(X, y), cv2.split(X, y)):
    assert np.array_equal(train1, train2)
    assert np.array_equal(test1, test2)

X, y = load_iris(return_X_y=True)
cv1 = StratifiedShuffleSplit(n_splits=5, train_size=0.5, test_size=0.2, random_state=0)
cv2 = skStratifiedShuffleSplit(n_splits=5, train_size=0.5, test_size=0.2, random_state=0)
for (train1, test1), (train2, test2) in zip(cv1.split(X, y), cv2.split(X, y)):
    assert np.array_equal(train1, train2)
    assert np.array_equal(test1, test2)