tfdeploy 0.0.1

Tiny, no-nonsense, self contained, TensorFlow inference
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
141
142
143
144
145
146
147
148
149

#[macro_use]
extern crate downcast_rs;
#[macro_use]
extern crate error_chain;
extern crate ndarray;
extern crate num_traits;
extern crate protobuf;

pub mod tfpb;
pub mod matrix;
pub mod ops;

use std::{fs, path, rc, str};
use std::collections::{HashMap, HashSet};
use ops::Op;

pub use matrix::Matrix;

error_chain!{
    foreign_links {
        Io(::std::io::Error);
        NdarrayShape(::ndarray::ShapeError);
        StrUtf8(::std::str::Utf8Error);
    }
}

pub struct Node {
    name: String,
    inputs: Vec<(rc::Rc<Node>, usize)>,
    op: Box<Op>,
}

impl Node {
    pub fn dump_eval_tree(&self) -> String {
        self._dump_eval_tree(0, &mut HashSet::new())
    }

    fn _dump_eval_tree(&self, depth: usize, dups: &mut HashSet<String>) -> String {
        let pad: String = ::std::iter::repeat("  ").take(depth).collect();
        let mut s = format!("{}{}\n", pad, self.name);
        for i in &self.inputs {
            s.push_str(&*format!("{}", i.0._dump_eval_tree(depth + 1, dups)));
        }
        s
    }
}

pub struct GraphAnalyser {
    graph: tfpb::GraphDef,
    op_builder: ops::OpBuilder,
    nodes: HashMap<String, rc::Rc<Node>>,
    outputs: HashMap<String, Vec<Matrix>>,
}

impl GraphAnalyser {
    pub fn new(graph: tfpb::GraphDef) -> GraphAnalyser {
        GraphAnalyser {
            graph,
            op_builder: ops::OpBuilder::new(),
            nodes: HashMap::new(),
            outputs: HashMap::new(),
        }
    }

    pub fn from_file<P: AsRef<path::Path>>(p: P) -> GraphAnalyser {
        Self::from_reader(fs::File::open(p).unwrap())
    }

    pub fn from_reader<R: ::std::io::Read>(mut r: R) -> GraphAnalyser {
        let loaded = ::protobuf::core::parse_from_reader::<::tfpb::graph::GraphDef>(&mut r)
            .unwrap();
        GraphAnalyser::new(loaded)
    }

    pub fn node_names(&self) -> Vec<&str> {
        self.graph.get_node().iter().map(|n| n.get_name()).collect()
    }

    pub fn get_node(&mut self, name: &str) -> Result<rc::Rc<Node>> {
        if !self.nodes.contains_key(name) {
            let node = self.make_node(name)?;
            self.nodes.insert(name.to_string(), rc::Rc::new(node));
        }
        Ok(self.nodes.get(name).unwrap().clone())
    }

    fn make_node(&mut self, name: &str) -> Result<Node> {
        let pbnode = self.graph
            .get_node()
            .iter()
            .find(|n| n.get_name() == name)
            .unwrap()
            .clone();
        Ok(Node {
            name: name.to_string(),
            inputs: pbnode
                .get_input()
                .iter()
                .map(|s| self.get_node(&s).map(|n| (n, 0)))
                .collect::<Result<_>>()?,
            op: self.op_builder.build(&pbnode)?,
        })
    }

    pub fn reset(&mut self) -> Result<()> {
        self.outputs.clear();
        Ok(())
    }

    pub fn set_value(&mut self, name: &str, value: Matrix) -> Result<()> {
        let node = self.get_node(name)?;
        if let Some(ph) = node.op.downcast_ref::<ops::trivial::Placeholder>() {
            ph.set(value);
            Ok(())
        } else {
            Err(format!("node {} is not a placeholder", name))?
        }
    }

    fn compute(&mut self, name: &str) -> Result<()> {
        if self.outputs.contains_key(name) {
            return Ok(());
        }
        let node: rc::Rc<Node> = self.get_node(name)?;
        let inputs: Vec<Matrix> = node.inputs
            .iter()
            .map(|i| {
                self.compute(&*i.0.name)?;
                Ok(self.outputs.get(&i.0.name).unwrap()[i.1].clone())
            })
            .collect::<Result<_>>()?;
        let outputs = node.op.eval(inputs)?;
        // println!("storing {} -> {:?}", name, outputs.iter().map(|m| m.shape()).collect::<Vec<_>>());
        self.outputs.insert(name.to_string(), outputs);
        Ok(())
    }

    pub fn eval(&mut self, name: &str) -> Result<Option<&Vec<Matrix>>> {
        self.compute(name)?;
        Ok(self.outputs.get(name))
    }

    pub fn take(&mut self, name: &str) -> Result<Vec<Matrix>> {
        self.compute(name)?;
        Ok(self.outputs.remove(name).ok_or(
            format!("{} does not exits", name),
        )?)
    }
}