# encoding: utf-8
"""
@author:  liaoxingyu
@contact: sherlockliao01@gmail.com
"""

import math

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import Parameter

from ..losses.loss_utils import one_hot


class Circle(nn.Module):
    def __init__(self, cfg, in_feat):
        super().__init__()
        self._num_classes = cfg.MODEL.HEADS.NUM_CLASSES
        self._s = cfg.MODEL.HEADS.SCALE
        self._m = cfg.MODEL.HEADS.MARGIN

        self.weight = Parameter(torch.Tensor(self._num_classes, in_feat))
        self.reset_parameters()

    def reset_parameters(self):
        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5))

    def forward(self, features, targets):
        sim_mat = F.linear(F.normalize(features), F.normalize(self.weight))
        alpha_p = F.relu(-sim_mat.detach() + 1 + self._m)
        alpha_n = F.relu(sim_mat.detach() + self._m)
        delta_p = 1 - self._m
        delta_n = self._m

        s_p = self._s * alpha_p * (sim_mat - delta_p)
        s_n = self._s * alpha_n * (sim_mat - delta_n)

        targets = one_hot(targets, self._num_classes)

        pred_class_logits = targets * s_p + (1.0 - targets) * s_n

        return pred_class_logits