Best Practices: Ray with Tensorflow¶
This document describes best practices for using the Ray core APIs with TensorFlow. Ray also provides higher-level utilities for working with Tensorflow, such as distributed training APIs (training tensorflow example), Tune for hyperparameter search (Tune tensorflow example), RLlib for reinforcement learning (RLlib tensorflow example).
Feel free to contribute if you think this document is missing anything.
Common Issues: Pickling¶
One common issue with TensorFlow2.0 is a pickling error like the following:
File "/home/***/venv/lib/python3.6/site-packages/ray/actor.py", line 322, in remote return self._remote(args=args, kwargs=kwargs) File "/home/***/venv/lib/python3.6/site-packages/ray/actor.py", line 405, in _remote self._modified_class, self._actor_method_names) File "/home/***/venv/lib/python3.6/site-packages/ray/function_manager.py", line 578, in export_actor_class "class": pickle.dumps(Class), File "/home/***/venv/lib/python3.6/site-packages/ray/cloudpickle/cloudpickle.py", line 1123, in dumps cp.dump(obj) File "/home/***/lib/python3.6/site-packages/ray/cloudpickle/cloudpickle.py", line 482, in dump return Pickler.dump(self, obj) File "/usr/lib/python3.6/pickle.py", line 409, in dump self.save(obj) File "/usr/lib/python3.6/pickle.py", line 476, in save f(self, obj) # Call unbound method with explicit self File "/usr/lib/python3.6/pickle.py", line 751, in save_tuple save(element) File "/usr/lib/python3.6/pickle.py", line 808, in _batch_appends save(tmp) File "/usr/lib/python3.6/pickle.py", line 496, in save rv = reduce(self.proto) TypeError: can't pickle _LazyLoader objects
To resolve this, you should move all instances of
import tensorflow into the Ray actor or function, as follows:
def create_model(): import tensorflow as tf ...
This issue is caused by side-effects of importing TensorFlow and setting global state.
Use Actors for Parallel Models¶
If you are training a deep network in the distributed setting, you may need to ship your deep network between processes (or machines). However, shipping the model is not always straightforward.
Avoid sending the Tensorflow model directly. A straightforward attempt to pickle a TensorFlow graph gives mixed results. Furthermore, creating a TensorFlow graph can take tens of seconds, and so serializing a graph and recreating it in another process will be inefficient.
It is recommended to replicate the same TensorFlow graph on each worker once at the beginning and then to ship only the weights between the workers.
Suppose we have a simple network definition (this one is modified from the TensorFlow documentation).
from tensorflow.keras import layers def create_keras_model(): import tensorflow as tf model = tf.keras.Sequential() # Adds a densely-connected layer with 64 units to the model: model.add(layers.Dense(64, activation="relu", input_shape=(32, ))) # Add another: model.add(layers.Dense(64, activation="relu")) # Add a softmax layer with 10 output units: model.add(layers.Dense(10, activation="softmax")) model.compile( optimizer=tf.keras.optimizers.RMSprop(0.01), loss=tf.keras.losses.categorical_crossentropy, metrics=[tf.keras.metrics.categorical_accuracy]) return model
It is strongly recommended you create actors to handle this. To do this, first initialize ray and define an Actor class:
import ray import numpy as np ray.init() def random_one_hot_labels(shape): n, n_class = shape classes = np.random.randint(0, n_class, n) labels = np.zeros((n, n_class)) labels[np.arange(n), classes] = 1 return labels # Use GPU wth # @ray.remote(num_gpus=1) @ray.remote class Network(object): def __init__(self): self.model = create_keras_model() self.dataset = np.random.random((1000, 32)) self.labels = random_one_hot_labels((1000, 10)) def train(self): history = self.model.fit(self.dataset, self.labels, verbose=False) return history.history def get_weights(self): return self.model.get_weights() def set_weights(self, weights): # Note that for simplicity this does not handle the optimizer state. self.model.set_weights(weights)
Then, we can instantiate this actor and train it on the separate process:
NetworkActor = Network.remote() result_object_id = NetworkActor.train.remote() ray.get(result_object_id)
We can then use
get_weights to move the weights of the neural network
around. This allows us to manipulate weights between different models running in parallel without shipping the actual TensorFlow graphs, which are much more complex Python objects.
NetworkActor2 = Network.remote() NetworkActor2.train.remote() weights = ray.get( [NetworkActor.get_weights.remote(), NetworkActor2.get_weights.remote()]) averaged_weights = [(layer1 + layer2) / 2 for layer1, layer2 in zip(weights, weights)] weight_id = ray.put(averaged_weights) [ actor.set_weights.remote(weight_id) for actor in [NetworkActor, NetworkActor2] ] ray.get([actor.train.remote() for actor in [NetworkActor, NetworkActor2]])
Lower-level TF Utilities¶
Given a low-level TF definition:
import tensorflow as tf import numpy as np x_data = tf.placeholder(tf.float32, shape=) y_data = tf.placeholder(tf.float32, shape=) w = tf.Variable(tf.random_uniform(, -1.0, 1.0)) b = tf.Variable(tf.zeros()) y = w * x_data + b loss = tf.reduce_mean(tf.square(y - y_data)) optimizer = tf.train.GradientDescentOptimizer(0.5) grads = optimizer.compute_gradients(loss) train = optimizer.apply_gradients(grads) init = tf.global_variables_initializer() sess = tf.Session()
To extract the weights and set the weights, you can use the following helper method.
import ray.experimental.tf_utils variables = ray.experimental.tf_utils.TensorFlowVariables(loss, sess)
TensorFlowVariables object provides methods for getting and setting the
weights as well as collecting all of the variables in the model.
Now we can use these methods to extract the weights, and place them back in the network as follows.
sess = tf.Session() # First initialize the weights. sess.run(init) # Get the weights weights = variables.get_weights() # Returns a dictionary of numpy arrays # Set the weights variables.set_weights(weights)
Note: If we were to set the weights using the
assign method like below,
each call to
assign would add a node to the graph, and the graph would grow
unmanageably large over time.
w.assign(np.zeros(1)) # This adds a node to the graph every time you call it. b.assign(np.zeros(1)) # This adds a node to the graph every time you call it.
TensorFlowVariables(output, sess=None, input_variables=None)¶
A class used to set and get weights for Tensorflow networks.
The tensorflow session used to run assignment.
Extracted variables from the loss or additional variables that are passed in.
Type: Dict[str, tf.Variable]
Placeholders for weights.
Type: Dict[str, tf.placeholders]
Nodes that assign weights.
Type: Dict[str, tf.Tensor]
Sets the current session used by the class.
Parameters: sess (tf.Session) – Session to set the attribute with.
Returns the total length of all of the flattened variables.
Returns: The length of all flattened variables concatenated.
Gets the weights and returns them as a flat array.
Returns: 1D Array containing the flattened weights.
Sets the weights to new_weights, converting from a flat array.
You can only set all weights in the network using this function, i.e., the length of the array must match get_flat_size.
Parameters: new_weights (np.ndarray) – Flat array containing weights.
Returns a dictionary containing the weights of the network.
Returns: Dictionary mapping variable names to their weights.
Sets the weights to new_weights.
Can set subsets of variables as well, by only passing in the variables you want to be set.
Parameters: new_weights (Dict) – Dictionary mapping variable names to their weights.
This may not work with tf.Keras.
TensorFlowVariables uses variable names to determine what
variables to set when calling
set_weights. One common issue arises when two
networks are defined in the same TensorFlow graph. In this case, TensorFlow
appends an underscore and integer to the names of variables to disambiguate
them. This will cause
TensorFlowVariables to fail. For example, if we have a
Network with a
import ray import tensorflow as tf class Network(object): def __init__(self): a = tf.Variable(1) b = tf.Variable(1) c = tf.add(a, b) sess = tf.Session() init = tf.global_variables_initializer() sess.run(init) self.variables = ray.experimental.tf_utils.TensorFlowVariables(c, sess) def set_weights(self, weights): self.variables.set_weights(weights) def get_weights(self): return self.variables.get_weights()
and run the following code:
a = Network() b = Network() b.set_weights(a.get_weights())
the code would fail. If we instead defined each network in its own TensorFlow graph, then it would work:
with tf.Graph().as_default(): a = Network() with tf.Graph().as_default(): b = Network() b.set_weights(a.get_weights())
This issue does not occur between actors that contain a network, as each actor
is in its own process, and thus is in its own graph. This also does not occur
Another issue to keep in mind is that
TensorFlowVariables needs to add new
operations to the graph. If you close the graph and make it immutable, e.g.
MonitoredTrainingSession the initialization will fail. To resolve
this, simply create the instance before you close the graph.