Merge pull request #68 from Enet4/export-saved-model

Load from saved model support

Author: adamcrume

examples/regression_savedmodel.py

@@ -0,0 +1,43 @@
+import tensorflow as tf
+from tensorflow.python.saved_model.builder import SavedModelBuilder
+from tensorflow.python.saved_model.signature_def_utils import build_signature_def
+from tensorflow.python.saved_model.signature_constants import REGRESS_METHOD_NAME
+from tensorflow.python.saved_model.tag_constants import TRAINING, SERVING
+from tensorflow.python.saved_model.utils import build_tensor_info
+
+x = tf.placeholder(tf.float32, name='x')
+y = tf.placeholder(tf.float32, name='y')
+
+w = tf.Variable(tf.random_uniform([1], -1.0, 1.0), name='w')
+b = tf.Variable(tf.zeros([1]), name='b')
+y_hat = w * x + b
+
+loss = tf.reduce_mean(tf.square(y_hat - y))
+optimizer = tf.train.GradientDescentOptimizer(0.5)
+train = optimizer.minimize(loss, name='train')
+
+init = tf.variables_initializer(tf.global_variables(), name='init')
+
+directory = 'examples/saved-regression-model'
+builder = SavedModelBuilder(directory)
+
+with tf.Session(graph=tf.get_default_graph()) as sess:
+    sess.run(init)
+
+    signature_inputs = {
+        "x": build_tensor_info(x),
+        "y": build_tensor_info(y)
+    }
+    signature_outputs = {
+        "out": build_tensor_info(y_hat)
+    }
+    signature_def = build_signature_def(
+        signature_inputs, signature_outputs,
+        REGRESS_METHOD_NAME)
+    builder.add_meta_graph_and_variables(
+        sess, [TRAINING, SERVING],
+        signature_def_map={
+            REGRESS_METHOD_NAME: signature_def
+        },
+        assets_collection=tf.get_collection(tf.GraphKeys.ASSET_FILEPATHS))
+    builder.save(as_text=False)

examples/regression_savedmodel.rs

@@ -0,0 +1,100 @@
+extern crate random;
+extern crate tensorflow;
+
+use random::Source;
+use std::error::Error;
+use std::result::Result;
+use std::path::Path;
+use std::process::exit;
+use tensorflow::Code;
+use tensorflow::Graph;
+use tensorflow::Session;
+use tensorflow::SessionOptions;
+use tensorflow::Status;
+use tensorflow::StepWithGraph;
+use tensorflow::Tensor;
+
+fn main() {
+    // Putting the main code in another function serves two purposes:
+    // 1. We can use the try! macro.
+    // 2. We can call exit safely, which does not run any destructors.
+    exit(match run() {
+        Ok(_) => 0,
+        Err(e) => {
+            println!("{}", e);
+            1
+        }
+    })
+}
+
+fn run() -> Result<(), Box<Error>> {
+    let export_dir = "examples/saved-regression-model"; // y = w * x + b
+    if !Path::new(export_dir).exists() {
+        return Err(Box::new(Status::new_set(Code::NotFound,
+                                            &format!("Run 'python regression_savedmodel.py' to generate \
+                                                      {} and try again.",
+                                                     export_dir))
+            .unwrap()));
+    }
+
+    // Generate some test data.
+    let w = 0.1;
+    let b = 0.3;
+    let num_points = 100;
+    let steps = 201;
+    let mut rand = random::default();
+    let mut x = Tensor::new(&[num_points as u64]);
+    let mut y = Tensor::new(&[num_points as u64]);
+    for i in 0..num_points {
+        x[i] = (2.0 * rand.read::<f64>() - 1.0) as f32;
+        y[i] = w * x[i] + b;
+    }
+
+    // Load the saved model exported by regression_savedmodel.py.
+    let mut graph = Graph::new();
+    let mut session = Session::from_saved_model(&SessionOptions::new(), 
+                                                &["train", "serve"],
+                                                &mut graph,
+                                                export_dir)?;
+    let op_x = graph.operation_by_name_required("x")?;
+    let op_y = graph.operation_by_name_required("y")?;
+    let op_train = graph.operation_by_name_required("train")?;
+    let op_w = graph.operation_by_name_required("w")?;
+    let op_b = graph.operation_by_name_required("b")?;
+
+    // Train the model (e.g. for fine tuning).
+    let mut train_step = StepWithGraph::new();
+    train_step.add_input(&op_x, 0, &x);
+    train_step.add_input(&op_y, 0, &y);
+    train_step.add_target(&op_train);
+    for _ in 0..steps {
+        try!(session.run(&mut train_step));
+    }
+
+    // Grab the data out of the session.
+    let mut output_step = StepWithGraph::new();
+    let w_ix = output_step.request_output(&op_w, 0);
+    let b_ix = output_step.request_output(&op_b, 0);
+    try!(session.run(&mut output_step));
+
+    // Check our results.
+    let w_hat: f32 = try!(output_step.take_output(w_ix)).data()[0];
+    let b_hat: f32 = try!(output_step.take_output(b_ix)).data()[0];
+    println!("Checking w: expected {}, got {}. {}",
+             w,
+             w_hat,
+             if (w - w_hat).abs() < 1e-3 {
+                 "Success!"
+             } else {
+                 "FAIL"
+             });
+    println!("Checking b: expected {}, got {}. {}",
+             b,
+             b_hat,
+             if (b - b_hat).abs() < 1e-3 {
+                 "Success!"
+             } else {
+                 "FAIL"
+             });
+    Ok(())
+}

src/session.rs

@@ -1,6 +1,8 @@
 use tf;
-use libc::c_int;
+use libc::{c_char, c_int};
+use std::ffi::CString;
 use std::marker;
+use std::path::Path;
 use std::ptr;
 use super::Code;
 use super::DataType;
@@ -32,6 +34,45 @@ impl Session {
         }
     }
 
+    /// Loads a session from an exported model.
+    pub fn from_saved_model<P: AsRef<Path>, Tag: AsRef<str>, Tags: IntoIterator<Item = Tag>>
+        (options: &SessionOptions,
+         tags: Tags,
+         graph: &mut Graph,
+         export_dir: P)
+         -> Result<Self> {
+        let mut status = Status::new();
+
+        let export_dir_cstr =
+            try!(export_dir.as_ref()
+                     .to_str()
+                     .and_then(|s| CString::new(s.as_bytes()).ok())
+                     .ok_or_else(|| invalid_arg!("Invalid export directory path")));
+
+        let tags_cstr: Vec<_> = try!(tags.into_iter()
+                                         .map(|t| CString::new(t.as_ref()))
+                                         .collect::<::std::result::Result<_, _>>()
+                                         .map_err(|_| invalid_arg!("Invalid tag name")));
+        // keeping tags_cstr to retain strings in memory
+        let tags_ptr: Vec<*const c_char> = tags_cstr.iter().map(|t| t.as_ptr()).collect();
+
+        let inner = unsafe {
+            tf::TF_LoadSessionFromSavedModel(options.inner,
+                                             ptr::null(),
+                                             export_dir_cstr.as_ptr(),
+                                             tags_ptr.as_ptr(),
+                                             tags_ptr.len() as c_int,
+                                             graph.inner(),
+                                             ptr::null_mut(),
+                                             status.inner())
+        };
+        if inner.is_null() {
+            Err(status)
+        } else {
+            Ok(Session { inner: inner })
+        }
+    }
+
     /// Closes the session.
     pub fn close(&mut self) -> Result<()> {
         let mut status = Status::new();
@@ -143,19 +184,19 @@ impl<'l> StepWithGraph<'l> {
                                     index: c_int,
                                     tensor: &'l Tensor<T>) {
         self.input_ports.push(tf::TF_Output {
-            oper: operation.inner(),
-            index: index,
-        });
+                                  oper: operation.inner(),
+                                  index: index,
+                              });
         self.input_tensors.push(tensor.inner);
     }
 
     /// Requests that an output is fetched from the graph after running this step.
     /// Returns an index that you can then use to fetch this output from the step after running it.
     pub fn request_output(&mut self, operation: &Operation, index: c_int) -> OutputToken {
         self.output_ports.push(tf::TF_Output {
-            oper: operation.inner(),
-            index: index,
-        });
+                                   oper: operation.inner(),
+                                   index: index,
+                               });
         self.output_tensors.push(ptr::null_mut());
         OutputToken { index: self.output_tensors.len() - 1 }
     }
@@ -172,13 +213,13 @@ impl<'l> StepWithGraph<'l> {
                                                  {}",
                                                 output_idx,
                                                 self.output_tensors.len()))
-                .unwrap());
+                               .unwrap());
         }
         if self.output_tensors[output_idx].is_null() {
             return Err(Status::new_set(Code::Unavailable,
                                        "Output not available. Either it was already taken, or \
                                         this step has not been sucessfully run yet.")
-                .unwrap());
+                               .unwrap());
         }
         let actual_data_type = self.output_data_type(output_idx).unwrap();
         if actual_data_type != T::data_type() {
@@ -260,13 +301,13 @@ mod tests {
         let y = {
             let mut nd = g.new_operation("Mul", "y").unwrap();
             nd.add_input(Output {
-                operation: &two,
-                index: 0,
-            });
+                             operation: &two,
+                             index: 0,
+                         });
             nd.add_input(Output {
-                operation: &x,
-                index: 0,
-            });
+                             operation: &x,
+                             index: 0,
+                         });
             nd.finish().unwrap()
         };
         let options = SessionOptions::new();