1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140

use super::DQNModelConfig;
use crate::{
    model::{ModelBase, SubModel},
    opt::{Optimizer, OptimizerConfig},
    util::OutDim,
};
use anyhow::Result;
use log::{info, trace};
use serde::{de::DeserializeOwned, Serialize};
use std::{marker::PhantomData, path::Path};
use tch::{nn, Device, Tensor};

#[allow(clippy::upper_case_acronyms)]
/// Represents value functions for DQN agents.
pub struct DQNModel<Q>
where
    Q: SubModel<Output = Tensor>,
    Q::Config: DeserializeOwned + Serialize + OutDim,
{
    device: Device,
    var_store: nn::VarStore,

    // Dimension of the output vector (equal to the number of actions).
    pub(super) out_dim: i64,

    // Action-value function
    q: Q,

    // Optimizer
    opt_config: OptimizerConfig,
    opt: Optimizer,

    phantom: PhantomData<Q>,
}

impl<Q> DQNModel<Q>
where
    Q: SubModel<Output = Tensor>,
    Q::Config: DeserializeOwned + Serialize + OutDim,
{
    pub fn build(config: DQNModelConfig<Q::Config>, device: Device) -> Self {
        let out_dim = config.q_config.as_ref().unwrap().get_out_dim();
        let opt_config = config.opt_config.clone();
        let var_store = nn::VarStore::new(device);
        let q = Q::build(&var_store, config.q_config.unwrap());

        Self::_build(device, out_dim, opt_config, q, var_store, None)
    }

    fn _build(
        device: Device,
        out_dim: i64,
        opt_config: OptimizerConfig,
        q: Q,
        mut var_store: nn::VarStore,
        var_store_src: Option<&nn::VarStore>,
    ) -> Self {
        // Optimizer
        let opt = opt_config.build(&var_store).unwrap();

        // Copy var_store
        if let Some(var_store_src) = var_store_src {
            var_store.copy(var_store_src).unwrap();
        }

        Self {
            device,
            out_dim,
            opt_config,
            var_store,
            opt,
            q,
            phantom: PhantomData,
        }
    }

    /// Outputs the action-value given an observation.
    pub fn forward(&self, x: &Q::Input) -> Tensor {
        let a = self.q.forward(&x);
        debug_assert_eq!(a.size().as_slice()[1], self.out_dim);
        a
    }
}

impl<Q> Clone for DQNModel<Q>
where
    Q: SubModel<Output = Tensor>,
    Q::Config: DeserializeOwned + Serialize + OutDim,
{
    fn clone(&self) -> Self {
        let device = self.device;
        let out_dim = self.out_dim;
        let opt_config = self.opt_config.clone();
        let var_store = nn::VarStore::new(device);
        let q = self.q.clone_with_var_store(&var_store);

        Self::_build(
            device,
            out_dim,
            opt_config,
            q,
            var_store,
            Some(&self.var_store),
        )
    }
}

impl<Q> ModelBase for DQNModel<Q>
where
    Q: SubModel<Output = Tensor>,
    Q::Config: DeserializeOwned + Serialize + OutDim,
{
    fn backward_step(&mut self, loss: &Tensor) {
        self.opt.backward_step(loss);
    }

    fn get_var_store_mut(&mut self) -> &mut nn::VarStore {
        &mut self.var_store
    }

    fn get_var_store(&self) -> &nn::VarStore {
        &self.var_store
    }

    fn save<T: AsRef<Path>>(&self, path: T) -> Result<()> {
        self.var_store.save(&path)?;
        info!("Save DQN model to {:?}", path.as_ref());
        let vs = self.var_store.variables();
        for (name, _) in vs.iter() {
            trace!("Save variable {}", name);
        }
        Ok(())
    }

    fn load<T: AsRef<Path>>(&mut self, path: T) -> Result<()> {
        self.var_store.load(&path)?;
        info!("Load DQN model from {:?}", path.as_ref());
        Ok(())
    }
}