local LSTMGateLayer = nerv.class('nerv.LSTMGateLayer', 'nerv.Layer')
-- NOTE: this is a full matrix gate

function LSTMGateLayer:__init(id, global_conf, layer_conf)
    nerv.Layer.__init(self, id, global_conf, layer_conf)
    self.param_type = layer_conf.param_type
    self:check_dim_len(-1, 1) --accept multiple inputs
    self:bind_params()
end

function LSTMGateLayer:bind_params()
    for i = 1, #self.dim_in do
        self["ltp" .. i] = self:find_param("ltp" .. i, self.lconf, self.gconf,
                                            nerv.LinearTransParam,
                                            {self.dim_in[i], self.dim_out[1]})
        if self.param_type[i] == 'D' then
            self["ltp" .. i].trans:diagonalize()
        end
    end
    self.bp = self:find_param("bp", self.lconf, self.gconf,
                                nerv.BiasParam, {1, self.dim_out[1]})
end

function LSTMGateLayer:init(batch_size)
    for i = 1, #self.dim_in do
        if self["ltp" .. i].trans:ncol() ~= self.bp.trans:ncol() then
            nerv.error("mismatching dimensions of linear transform and bias paramter")
        end
        if self.dim_in[i] ~= self["ltp" .. i].trans:nrow() then
            nerv.error("mismatching dimensions of linear transform parameter and input")
        end
        self["ltp"..i]:train_init()
    end
    
    if self.dim_out[1] ~= self.ltp1.trans:ncol() then
        nerv.error("mismatching dimensions of linear transform parameter and output")
    end
    self.bp:train_init()
    self.err_bakm = self.gconf.cumat_type(batch_size, self.dim_out[1])
end

function LSTMGateLayer:batch_resize(batch_size)
    if self.err_m:nrow() ~= batch_size then
        self.err_bakm = self.gconf.cumat_type(batch_size, self.dim_out[1])
    end
end

function LSTMGateLayer:propagate(input, output)
    -- apply linear transform
    output[1]:mul(input[1], self.ltp1.trans, 1.0, 0.0, 'N', 'N')
    for i = 2, #self.dim_in do
        output[1]:mul(input[i], self["ltp" .. i].trans, 1.0, 1.0, 'N', 'N')
    end
    -- add bias
    output[1]:add_row(self.bp.trans, 1.0)
    output[1]:sigmoid(output[1])
end

function LSTMGateLayer:back_propagate(bp_err, next_bp_err, input, output)
    self.err_bakm:sigmoid_grad(bp_err[1], output[1])
    for i = 1, #self.dim_in do
        next_bp_err[i]:mul(self.err_bakm, self["ltp" .. i].trans, 1.0, 0.0, 'N', 'T')
    end
end

function LSTMGateLayer:update(bp_err, input, output)
    self.err_bakm:sigmoid_grad(bp_err[1], output[1])
    for i = 1, #self.dim_in do
        self["ltp" .. i]:update_by_err_input(self.err_bakm, input[i])
        if self.param_type[i] == 'D' then
            self["ltp" .. i].trans:diagonalize()
        end
    end
    self.bp:update_by_gradient(self.err_bakm:colsum())
end

function LSTMGateLayer:get_params()
    local pr = nerv.ParamRepo({self.bp}, self.loc_type)
    for i = 1, #self.dim_in do
        pr:add(self["ltp" .. i].id, self["ltp" .. i])
    end
    return pr
end