def __init__(self, bins=10, momentum=0.0, mu=0.02): super(GHMRLoss, self).__init__() self.bins = bins self.momentum = momentum self.mu = mu edges_left = np.array([float(x) / bins for x in range(bins)], dtype=np.float32) self.edges_left = Tensor(edges_left.reshape((bins, 1, 1, 1, 1))) edges_right = np.array([float(x) / bins for x in range(1, bins + 1)], dtype=np.float32) edges_right[-1] += 1e-4 self.edges_right = Tensor(edges_right.reshape((bins, 1, 1, 1, 1))) if momentum >= 0: self.acc_sum = Parameter(initializer(0, [bins], mstype.float32)) self.abs = ops.Abs() self.sqrt = ops.Sqrt() self.cast = ops.Cast() self.select = ops.Select() self.reshape = ops.Reshape() self.reduce_sum = ops.ReduceSum() self.max = ops.Maximum() self.less = ops.Less() self.equal = ops.Equal() self.greater = ops.Greater() self.logical_and = ops.LogicalAnd() self.greater_equal = ops.GreaterEqual() self.zeros_like = ops.ZerosLike() self.expand_dims = ops.ExpandDims()
def construct(self, x, seq_lengths): """Defines the ReverseSequence operator computation performed.""" batch_size = x.shape[self.batch_dim] max_seq_len = x.shape[self.seq_dim] seq_lens_type = seq_lengths.dtype back = ops.Sub()(seq_lengths, ops.OnesLike()(seq_lengths)) batch_idx = self.make_shape((batch_size, max_seq_len), seq_lens_type, 0) forward_idx = self.make_shape((batch_size, max_seq_len), seq_lens_type, 1) back = back.view(-1, 1) reverse_idx = ops.Sub()(back, forward_idx) condition = ops.Less()(reverse_idx, ops.ZerosLike()(reverse_idx)) reverse_idx = ops.Select()(condition, forward_idx, reverse_idx) reverse_idx = ops.ExpandDims()(reverse_idx, 2) batch_idx = ops.ExpandDims()(batch_idx, 2) if self.batch_dim > self.seq_dim: batch_idx = ops.Transpose()(batch_idx, (1, 0, 2)) reverse_idx = ops.Transpose()(reverse_idx, (1, 0, 2)) x = ops.Transpose()(x, (1, 0, 2)) start_indices = ops.Concat(2)((batch_idx, reverse_idx)) output = ops.GatherNd()(x, start_indices) return output
def __init__(self, alpha=2, beta=4): super(FocalLoss, self).__init__() self.alpha = alpha self.beta = beta self.pow = ops.Pow() self.log = ops.Log() self.select = ops.Select() self.equal = ops.Equal() self.less = ops.Less() self.cast = ops.Cast() self.fill = ops.Fill() self.dtype = ops.DType() self.shape = ops.Shape() self.reduce_sum = ops.ReduceSum()
def hardshrink(input, lambd=0.5): great_lambd = ops.Greater()(input, lambd) less_neg_lambd = ops.Less()(input, lambd) cond = ops.logical_or(great_lambd, less_neg_lambd) return ops.Select()(cond, input, ops.scalar_to_tensor(0.0))