FateWithShellmiaoComment/Fate/python/federatedml/nn/backend/utils/rng.py

import random
from fate_arch.session import computing_session
import numpy as np
from federatedml.secureprotol.paillier_tensor import PaillierTensor


BITS = 10
MIXED_RATE = 0.5


class RandomNumberGenerator(object):
    def __init__(self):
        self.lower_bound = -2 ** BITS
        self.upper_bound = 2 ** BITS

    @staticmethod
    def get_size_by_shape(shape):
        size = 1
        for dim in shape:
            size *= dim

        return size

    def generate_random_number_1d(
            self,
            size,
            mixed_rate=MIXED_RATE,
            keep=None):
        if keep is not None:
            ret = [0] * size
            for i in range(size):
                if keep[i]:
                    rng = random.SystemRandom().uniform(
                        self.lower_bound,
                        self.upper_bound) if np.random.rand() < mixed_rate else np.random.uniform(
                        self.lower_bound,
                        self.upper_bound)
                    ret[i] = rng

            return np.array(ret)[keep]
        else:
            return [
                random.SystemRandom().uniform(
                    self.lower_bound,
                    self.upper_bound) if np.random.rand() < mixed_rate else np.random.uniform(
                    self.lower_bound,
                    self.upper_bound) for _ in range(size)]

    def generate_random_number(
            self,
            shape=None,
            mixed_rate=MIXED_RATE,
            keep=None):
        if keep is not None:
            size = self.get_size_by_shape(keep.shape)
            return self.generate_random_number_1d(
                size, mixed_rate=mixed_rate, keep=keep)
        else:
            size = self.get_size_by_shape(shape)
            return np.reshape(
                self.generate_random_number_1d(
                    size, mixed_rate=mixed_rate), shape)

    def fast_generate_random_number(
            self,
            shape,
            partition=10,
            mixed_rate=MIXED_RATE,
            keep_table=None):
        if keep_table:
            tb = keep_table.mapValues(
                lambda keep_array: self.generate_random_number(
                    keep=keep_array, mixed_rate=mixed_rate))
            return PaillierTensor(tb)
        else:
            tb = computing_session.parallelize(
                [None for _ in range(shape[0])], include_key=False, partition=partition)

            tb = tb.mapValues(lambda val: self.generate_random_number(
                shape[1:], mixed_rate=mixed_rate))

            return PaillierTensor(tb)