Source code for nengo_dl.tensor_graph

from collections import OrderedDict
import datetime
import logging
import time
import warnings

from nengo import Connection, Process
from nengo.builder.operator import TimeUpdate, SimPyFunc
from nengo.builder.processes import SimProcess
from nengo.exceptions import SimulationError
from nengo.neurons import Direct
import tensorflow as tf

from nengo_dl import (builder, graph_optimizer, signals, utils, tensor_node,
                      nengo_version)

logger = logging.getLogger(__name__)


[docs]class TensorGraph(object): """Manages the construction of the TensorFlow symbolic computation graph. Parameters ---------- model : :class:`~nengo:nengo.builder.Model` pre-built Nengo model describing the network to be simulated dt : float length of a simulator timestep, in seconds step_blocks : int controls how many simulation steps run each time the graph is executed (affects memory usage and graph construction time) unroll_simulation : bool if True, unroll simulation loop by explicitly building each iteration (up to ``step_blocks``) into the computation graph. if False, use a symbolic loop, which is more general and produces a simpler graph, but is likely to be slower to simulate dtype : ``tf.DType`` floating point precision to use for simulation minibatch_size : int the number of simultaneous inputs that will be passed through the network device : None or ``"/cpu:0"`` or ``"/gpu:[0-n]"`` device on which to execute computations (if None then uses the default device as determined by Tensorflow) """ def __init__(self, model, dt, step_blocks, unroll_simulation, dtype, minibatch_size, device): self.model = model self.dt = dt self.step_blocks = step_blocks self.unroll_simulation = unroll_simulation self.dtype = dtype self.minibatch_size = minibatch_size self.device = device # find invariant inputs (nodes that don't receive any input other # than the simulation time). we'll compute these outside the simulation # and feed in the result. if self.model.toplevel is None: self.invariant_inputs = [] else: self.invariant_inputs = [n for n in self.model.toplevel.all_nodes if n.size_in == 0 and not isinstance(n, tensor_node.TensorNode)] # filter unused operators # remove TimeUpdate because it is executed as part of the simulation # loop, not part of the step plan. remove input nodes because they # are executed outside the simulation. node_processes = [n.output for n in self.invariant_inputs if isinstance(n.output, Process)] operators = [ op for op in self.model.operators if not ( isinstance(op, TimeUpdate) or (isinstance(op, SimPyFunc) and op.x is None) or (isinstance(op, SimProcess) and op.input is None and op.process in node_processes))] # mark trainable signals mark_signals(model) logger.info("Initial plan length: %d", len(operators)) utils.print_and_flush("Optimizing graph", end="") start = time.time() # group mergeable operators if nengo_version < (2, 4, 0): plan = graph_optimizer.greedy_planner(operators) else: plan = graph_optimizer.transitive_planner(operators) # order signals/operators to promote contiguous reads sigs, self.plan = graph_optimizer.order_signals(plan, n_passes=10) # create base arrays and map Signals to TensorSignals (views on those # base arrays) self.base_arrays_init, self.sig_map = graph_optimizer.create_signals( sigs, self.plan, float_type=dtype.as_numpy_dtype, minibatch_size=self.minibatch_size) print("\rOptimization completed in %s " % datetime.timedelta(seconds=int(time.time() - start))) logger.info("Optimized plan length: %d", len(self.plan)) logger.info("Number of base arrays: %d", len(self.base_arrays_init))
[docs] def build(self, rng): """Constructs a new graph to simulate the model. Parameters ---------- rng : :class:`~numpy:numpy.random.RandomState` the Simulator's random number generator """ self.graph = tf.Graph() self.signals = signals.SignalDict(self.sig_map, self.dtype, self.minibatch_size) self.target_phs = {} self.losses = {} self.optimizers = {} with self.graph.as_default(), tf.device(self.device): # make sure indices are loaded for all probe signals (they won't # have been loaded if this signal is only accessed as part of a # larger block during the simulation) for p in self.model.probes: self.sig_map[self.model.sig[p]["in"]].load_indices() # create this constant once here so we don't end up creating a new # dt constant in each operator self.signals.dt = tf.constant(self.dt, self.dtype) self.signals.dt_val = self.dt # store the actual value as well # create base arrays self.base_vars = [] for k, (v, trainable) in self.base_arrays_init.items(): unique_idx = 0 duplicate = True while duplicate: name = "%s_%s_%s_%s" % ( v.dtype, "_".join(str(x) for x in v.shape), trainable, unique_idx) if any([name in x.name for x in ( tf.global_variables() if trainable else tf.local_variables())]): unique_idx += 1 else: try: self.graph.get_tensor_by_name(name + ":0") unique_idx += 1 except KeyError: duplicate = False # if trainable: # # trainable signal, so create Variable # with tf.variable_scope("base_vars", reuse=False): # var = tf.get_variable( # name, initializer=tf.constant_initializer(v), # dtype=v.dtype, shape=v.shape, trainable=True) # else: # var = tf.placeholder(tf.as_dtype(v.dtype), shape=v.shape, # name=name) if trainable: with tf.variable_scope("trainable_vars", reuse=False): var = tf.get_variable( name, initializer=tf.constant_initializer(v), dtype=v.dtype, shape=v.shape, trainable=True) else: with tf.variable_scope("local_vars", reuse=False): var = tf.get_local_variable( name, initializer=tf.constant_initializer(v), dtype=v.dtype, shape=v.shape, trainable=False) self.base_vars += [var] logger.debug("created base arrays") logger.debug([str(x) for x in self.base_vars]) # set up invariant inputs self.build_inputs(rng) # pre-build stage for ops in self.plan: with self.graph.name_scope(utils.sanitize_name( builder.Builder.builders[type(ops[0])].__name__)): builder.Builder.pre_build(ops, self.signals, rng) # build stage self.build_loop() # ops for initializing variables (will be called by simulator) self.trainable_init_op = tf.variables_initializer( tf.trainable_variables()) self.local_init_op = tf.local_variables_initializer() # note: the only non-trainable global variables should be those # created inside TensorNodes self.global_init_op = tf.variables_initializer( [v for v in tf.global_variables() if v not in tf.trainable_variables()])
[docs] def build_step(self): """Build the operators that execute a single simulation timestep into the graph. Returns ------- probe_tensors : list of ``tf.Tensor`` the Tensor objects representing the data required for each model Probe side_effects : list of ``tf.Tensor`` the output Tensors of computations that may have side-effects (e.g., :class:`~nengo:nengo.Node` functions), meaning that they must be executed each time step even if their output doesn't appear to be used in the simulation """ # build operators side_effects = [] # manually build TimeUpdate. we don't include this in the plan, # because loop variables (`step`) are (semi?) pinned to the CPU, which # causes the whole variable to get pinned to the CPU if we include # `step` as part of the normal planning process. self.signals.time = tf.cast(self.signals.step, self.dtype) * self.signals.dt # build operators for ops in self.plan: with self.graph.name_scope(utils.sanitize_name( builder.Builder.builders[type(ops[0])].__name__)): outputs = builder.Builder.build(ops, self.signals) if outputs is not None: side_effects += outputs # TODO: better solution to avoid the forced_copy # we need to make sure that probe reads occur before the # probe value is overwritten on the next timestep. however, # just blocking on the sliced value (probe_tensor) doesn't # work, because slices of variables don't perform a # copy, so the slice can be "executed" and then the value # overwritten before the tensorarray write occurs. what we # really want to do is block until the probe_arrays.write # happens, but you can't block on probe_arrays (and blocking on # probe_array.flow doesn't work, although I think it should). # so by adding the copy here and then blocking on the copy, we make # sure that the probe value is read before it can be overwritten. logger.debug("collecting probe tensors") probe_tensors = [ self.signals.gather(self.sig_map[self.model.sig[p]["in"]], force_copy=True) for p in self.model.probes] logger.debug("=" * 30) logger.debug("build_step complete") logger.debug("probe_tensors %s", [str(x) for x in probe_tensors]) logger.debug("side_effects %s", [str(x) for x in side_effects]) return probe_tensors, side_effects
[docs] def build_loop(self): """Build simulation loop. Loop can be constructed using the ``tf.while_loop`` architecture, or explicitly unrolled. Unrolling increases graph construction time and memory usage, but increases simulation speed. """ def loop_condition(step, stop, *_): return step < stop def loop_body(step, stop, loop_i, probe_arrays, base_vars): self.signals.bases = OrderedDict( [(k, v) for k, v in zip(self.base_arrays_init.keys(), base_vars)]) # note: nengo step counter is incremented at the beginning of # the timestep step += 1 self.signals.step = step # fill in invariant input data for n in self.invariant_ph: self.signals.scatter( self.sig_map[self.model.sig[n]["out"]], self.invariant_ph[n][loop_i]) # build the operators for a single step # note: we tie things to the `loop_i` variable so that we can be # sure the other things we're tying to the simulation step (side # effects and probes) from the previous timestep are executed # before the next step starts with self.graph.control_dependencies([loop_i]): probe_tensors, side_effects = self.build_step() # copy probe data to array for i, p in enumerate(probe_tensors): probe_arrays[i] = probe_arrays[i].write(loop_i, p) # need to make sure that any operators that could have side # effects run each timestep, so we tie them to the loop increment. # we also need to make sure that all the probe reads happen before # those values get overwritten on the next timestep with self.graph.control_dependencies(side_effects + probe_tensors): loop_i += 1 base_vars = tuple(self.signals.bases.values()) return step, stop, loop_i, probe_arrays, base_vars self.step_var = tf.placeholder(tf.int32, shape=(), name="step") self.stop_var = tf.placeholder(tf.int32, shape=(), name="stop") loop_i = tf.constant(0) probe_arrays = [ tf.TensorArray( self.signals.dtype, clear_after_read=False, size=0 if self.step_blocks is None else self.step_blocks, dynamic_size=self.step_blocks is None) for _ in self.model.probes] # build simulation loop loop_vars = ( self.step_var, self.stop_var, loop_i, probe_arrays, tuple(x._ref() if isinstance(x, tf.Variable) else x for x in self.base_vars)) if self.unroll_simulation: for n in range(self.step_blocks): logger.debug("BUILDING ITERATION %d", n) with self.graph.name_scope("iteration_%d" % n): loop_vars = loop_body(*loop_vars) else: # TODO: get parallel iterations working? nengo simulations are # pretty serial though, so I'm not sure how much benefit we would # get (and it seems non-trivial to get working correctly) loop_vars = tf.while_loop( loop_condition, loop_body, loop_vars=loop_vars, parallel_iterations=1, back_prop=True) self.end_base_arrays = loop_vars[4] self.probe_arrays = [] for p in loop_vars[3]: x = p.stack() if self.step_blocks is not None: x.set_shape([self.step_blocks] + x.get_shape().as_list()[1:]) self.probe_arrays += [x] # note: we need to make sure the final base array updates get computed, # even if they aren't being read by anything, because they may be # being read on the next `_run_steps` call. the `tf.while_loop` # enter/exit logic takes care of that on its own, so we only need to # do this for the unrolled case with tf.control_dependencies(self.end_base_arrays if self.unroll_simulation else []): self.steps_run = tf.identity(loop_vars[2])
[docs] def build_inputs(self, rng): """Sets up the inputs in the model (which will be computed outside of Tensorflow and fed in each simulation block). Parameters ---------- rng : :class:`~numpy:numpy.random.RandomState` the Simulator's random number generator """ self.invariant_funcs = {} self.invariant_ph = {} for n in self.invariant_inputs: if self.model.sig[n]["out"] in self.sig_map: # make sure the indices for this input are loaded into # TensorFlow (they may not be, if the output of this node is # only read as part of a larger block during the simulation) self.sig_map[self.model.sig[n]["out"]].load_indices() # set up a placeholder input for this node self.invariant_ph[n] = tf.placeholder( self.dtype, (self.step_blocks, n.size_out, self.minibatch_size)) # build the node functions, which will be called offline to # generate the input values if isinstance(n.output, Process): self.invariant_funcs[n] = n.output.make_step( (n.size_in,), (n.size_out,), self.dt, n.output.get_rng(rng)) elif n.size_out > 0: self.invariant_funcs[n] = utils.align_func( (n.size_out,), self.dtype)(n.output) else: self.invariant_funcs[n] = n.output
[docs] def build_optimizer(self, optimizer, targets, objective): """Adds elements into the graph to execute the given optimizer. Parameters ---------- optimizer : ``tf.train.Optimizer`` instance of a Tensorflow optimizer class targets : tuple of :class:`~nengo:nengo.Probe` the Probes corresponding to the output signals being optimized objective : ``"mse"`` or callable the objective to be minimized. passing ``"mse"`` will train with mean squared error. a custom function ``f(output, target) -> loss`` can be passed that consumes the actual output and target output for a probe in ``targets`` and returns a ``tf.Tensor`` representing the scalar loss value for that Probe (loss will be averaged across Probes). """ with self.graph.as_default(), tf.device(self.device): loss = self.build_loss(objective, targets) key = (optimizer, targets, objective) if key not in self.optimizers: # create optimizer operator try: opt_op = optimizer.minimize( loss, var_list=tf.trainable_variables()) except ValueError: raise SimulationError( "Network graph contains non-differentiable elements") # get any new variables created by optimizer (so they can be # initialized) opt_slots_init = tf.variables_initializer( [optimizer.get_slot(v, name) for v in tf.trainable_variables() for name in optimizer.get_slot_names()]) self.optimizers[key] = (opt_op, opt_slots_init) return self.optimizers[key]
[docs] def build_loss(self, objective, targets): """Adds elements into the graph to compute the given objective. Parameters ---------- objective : ``"mse"`` or callable the objective used to compute loss. passing ``"mse"`` will use mean squared error. a custom function ``f(output, target) -> loss`` can be passed that consumes the actual output and target output for a probe in ``targets`` and returns a ``tf.Tensor`` representing the scalar loss value for that Probe (loss will be averaged across Probes). targets : tuple of :class:`~nengo:nengo.Probe` the Probes corresponding to target values in objective """ if isinstance(targets, list): targets = tuple(targets) if (objective, targets) in self.losses: return self.losses[(objective, targets)] with self.graph.as_default(), tf.device(self.device): loss = [] for p in targets: probe_index = self.model.probes.index(p) # create a placeholder for the target values if p not in self.target_phs: self.target_phs[p] = tf.placeholder( self.dtype, (self.step_blocks, p.size_in, self.minibatch_size), name="targets") # compute loss if objective == "mse": loss += [tf.reduce_mean(tf.square( self.target_phs[p] - self.probe_arrays[probe_index]))] elif callable(objective): # move minibatch dimension back to the front x = tf.transpose(self.probe_arrays[probe_index], (2, 0, 1)) t = tf.transpose(self.target_phs[p], (2, 0, 1)) loss += [objective(x, t)] else: raise NotImplementedError # average loss across probes (note: this will also average across # the output of `objective` if it doesn't return a scalar) loss = tf.reduce_mean(loss) self.losses[(objective, targets)] = loss return loss
[docs]def mark_signals(model): """Mark all the signals in ``model`` according to whether they represent trainable parameters of the model (parameters that can be optimized by deep learning methods). Trainable parameters include connection weights, ensemble encoders, and neuron biases. Unless one of those signals is targeted by a Nengo learning rule (otherwise the learning rule update conflicts with the deep learning optimization). Parameters ---------- model : class:`~nengo:nengo.builder.Model` built Nengo model """ if model.toplevel is None: warnings.warn("No top-level network in model") else: # encoders and biases are trainable for ens in model.toplevel.all_ensembles: model.sig[ens]["encoders"].trainable = True model.sig[ens]["encoders"].minibatched = False if not isinstance(ens.neuron_type, Direct): model.sig[ens.neurons]["bias"].trainable = True model.sig[ens.neurons]["bias"].minibatched = False # connection weights are trainable for conn in model.toplevel.all_connections: # note: this doesn't include probe connections, since they aren't # added to the network # TODO: should we disable training on connections to learning # rules? model.sig[conn]["weights"].trainable = True model.sig[conn]["weights"].minibatched = False # parameters can't be modified by an online Nengo learning rule # and offline training at the same time. (it is possible in theory, # but it complicates things a lot and is probably not a common # use case). we also make those signals minibatched (they # wouldn't be normally), because we want to be able to learn # independently in each minibatch for conn in model.toplevel.all_connections: rule = conn.learning_rule if rule is not None: if isinstance(rule, dict): rule = list(rule.values()) elif not isinstance(rule, list): rule = [rule] for r in rule: if r.modifies == "weights" or r.modifies == "decoders": model.sig[conn]["weights"].trainable = False model.sig[conn]["weights"].minibatched = True elif r.modifies == "encoders": model.sig[conn.post_obj]["encoders"].trainable = False model.sig[conn.post_obj]["encoders"].minibatched = True else: raise NotImplementedError # the connections to connection probes are not trainable, but # also not minibatched probe_seeds = [model.seeds[p] for p in model.probes] for obj, seed in model.seeds.items(): if isinstance(obj, Connection) and seed in probe_seeds: model.sig[obj]["weights"].trainable = False model.sig[obj]["weights"].minibatched = False # fill in defaults for all other signals # signals are not trainable by default, and views take on the properties # of their bases for op in model.operators: for sig in op.all_signals: if not hasattr(sig.base, "trainable"): sig.base.trainable = False if not hasattr(sig.base, "minibatched"): sig.base.minibatched = not sig.base.trainable if not hasattr(sig, "trainable"): sig.trainable = sig.base.trainable if not hasattr(sig, "minibatched"): sig.minibatched = sig.base.minibatched for p in model.probes: sig = model.sig[p]["in"] if not hasattr(sig, "trainable"): sig.trainable = sig.base.trainable if not hasattr(sig, "minibatched"): sig.minibatched = sig.base.minibatched