def process_node(self, env, node): # helpful functions def select_min(x, y): if x is None: return y if y is None: return x return tensor.minimum(x, y) def select_max(x, y): if x is None: return y if y is None: return x return tensor.maximum(x, y) def sanitize(x): if x is None: return None else: return tensor.as_tensor_variable(x) shape_of = node.env.shape_feature.shape_of # 1. Initialization of variables # Note 1) We do not actually care about outputs representing shared # variables (those have no intermediate values) so it is safer to # ignore them and not change them in any way. To simplify the # optimizations I construct the variable ``c_outs`` ( that counts # outputs up to those we care) and the list ``init_l`` which for any # output we care says the length of its initial state. Note that # defining ``init_l`` for mit_mot sequences is a bit trickier but # it is safe to set it to 0 op = node.op c_outs = op.n_mit_mot + op.n_mit_sot + op.n_sit_sot + op.n_nit_sot init_l = [0 for x in xrange(op.n_mit_mot)] init_l += [abs(numpy.min(v)) for v in op.tap_array[op.n_mit_mot :]] init_l += [0 for x in xrange(op.n_nit_sot)] # 2. Check the clients of each output and see for how many steps # does scan need to run # This comparison checks if there is any uncounted output, which # can only be an output corresponding to a shared variable # 2.1 Initialize # global_nsteps is a dictionary having two fields ( 'real' deals # with int values, 'sym' with symbolic ones) or None # given that a scan op has k outputs o_1, .. o_k and each # output has n_j clients c_1^1, c_1^2, .. c_1^{n_1}, c_2^1, .., # global_nsteps is None if any of the clients is different # from a subtensor or its real and sym field equal to # max(c_i_j.idx_list[0].stop), meaning store up to which maximal # index(step) for any output scan actually needs to compute # In other words n_steps should be equal to this maximal ! # Note: if we have a shared variable that gets updated at every step # of the loop, reducing the number of steps will affect the the # value of the shared variable after the loop so we need not to # change the number of steps in that case. To do this we set # global_nsteps to None which is seen as a flag that nothing needs # to be done if len(node.outputs) <= c_outs: global_nsteps = {"real": -1, "sym": []} else: global_nsteps = None # Keeps track of the original slices that each client represent slices = [None for o in node.outputs] # A list for each output indicating how many intermediate values # should be stored. If negative it means none of the intermediate # values (i.e. the output can be removed since it is not used # afterwards in the computations), if 0 it means that all # intermediate values are required, otherwise is up to that number # of intermediate values # Note that for mit_mot outputs and shared outputs we can not change # the number of intermediate steps stored without affecting the # result of the op store_steps = [0 for o in xrange(op.n_mit_mot)] store_steps += [-1 for o in node.outputs[op.n_mit_mot : c_outs]] # Flag that says if an input has changed and we need to do something # or not flag_store = False # 2.2 Loop over the clients for i, out in enumerate(node.outputs[:c_outs]): # look at all its clients slices[i] = [] for cl, _ in out.clients: # 2.1 outputs of the function # => output needs all its intermediate values if type(cl) == str: # if the node is actually an output, then # we need to store the entire thing global_nsteps = None slices[i] = None break # 2.2 non-subtensor nodes # => output needs all its intermediate values elif not isinstance(cl.op, tensor.basic.Subtensor): global_nsteps = None slices[i] = None break # 2.3 subtensor nodes # => output might need to store just a subset of its values else: # 2.3.1 extract idx list of subtensor this_slice = tensor.basic.get_idx_list(cl.inputs, cl.op.idx_list) if this_slice == None: # if unable to extract idx_list # => outputs needs all its intermediate values global_nsteps = None slices[i] = None break # 2.3.2 extract the begin/end of the first dimension if i > op.n_mit_mot: try: length = shape_of[out][0] except Exception: length = node.inputs[0] + init_l[i] else: try: length = shape_of[out][0] except Exception: length = out.shape[0] cf_slice = tensor.basic.get_canonical_form_slice(this_slice[0], length) slices[i] += [(cf_slice, this_slice)] if isinstance(this_slice[0], slice) and this_slice[0].stop is None: global_nsteps = None break if isinstance(cf_slice[0], slice): stop = tensor.basic.extract_constant(cf_slice[0].stop) else: stop = tensor.basic.extract_constant(cf_slice[0]) + 1 if stop == sys.maxint or stop == length: stop = None else: # there is a **gotcha** here ! Namely, scan returns an # array that contains the initial state of the output # as well. Which means that if have a initial state of # length 3, and you look for 5 steps you get an output # y of length 8. If you only use y[:5], this does not # mean that you only need to loop for 5 steps but # actually only for 2 steps ( the first 3 are the # initial state) stop = stop - init_l[i] # 2.3.3 we might get away with less number of steps if stop is not None and global_nsteps is not None: # yes if it is a tensor if isinstance(stop, tensor.Variable): global_nsteps["sym"] += [stop] # not if it is maxint elif type(stop) is int and stop == sys.maxint: global_nsteps = None # yes if it is a int k, 0 < k < maxint elif type(stop) is int and global_nsteps["real"] < stop: global_nsteps["real"] = stop # yes if it is a int k, 0 < k < maxint elif type(stop) is int and stop > 0: pass # not otherwise else: global_nsteps = None # 2.3. Analyze global_nsteps to figure out for how many steps scan # needs to iterate if global_nsteps is not None: nw_steps = node.inputs[0] # there are some symbolic tensors that limit the number of # steps if len(global_nsteps["sym"]) == 0: sym_steps = None else: sym_steps = global_nsteps["sym"][0] for c in global_nsteps["sym"][1:]: sym_steps = tensor.maximum(sym_steps, c) if global_nsteps["real"] >= 0: real_steps = global_nsteps["real"] else: real_steps = None nw_steps = select_min(select_max(sym_steps, real_steps), node.inputs[0]) else: nw_steps = node.inputs[0] global_nsteps = None # 2.4 Loop over the clients again now looking just to see how many # intermediate steps to store for i, out in enumerate(node.outputs[:c_outs]): # look at all its clients for cl, _ in out.clients: if type(cl) == str: store_steps[i] = 0 break elif not isinstance(cl.op, tensor.basic.Subtensor): store_steps[i] = 0 break else: this_slice = tensor.basic.get_idx_list(cl.inputs, cl.op.idx_list) if this_slice == None: store_steps[i] = 0 break if isinstance(this_slice[0], slice) and this_slice[0].start is None: store_steps[i] = 0 break if i > op.n_mit_mot: length = node.inputs[0] + init_l[i] else: try: length = shape_of[out][0] except Exception: length = out.shape[0] cf_slice = tensor.basic.get_canonical_form_slice(this_slice[0], length) if isinstance(cf_slice[0], slice): start = tensor.basic.extract_constant(cf_slice[0].start) else: start = tensor.basic.extract_constant(cf_slice[0]) if start == 0 or store_steps[i] == 0: store_steps[i] = 0 else: pval = select_max(nw_steps - start + init_l[i], init_l[i]) if store_steps[i] != -1: pval = select_max(pval, store_steps[i]) store_steps[i] = pval flag_store = True orphane_outs = [i for i, x in enumerate(store_steps) if (type(x) is int) and (x < 0)] flag_store = flag_store or (len(orphane_outs) > 0) # 3. is there anything to change ? if flag_store or global_nsteps is not None: # 3.1 initialize inputs for the new scan old_outputs = [] nw_inputs = list(node.inputs) nw_inputs[0] = nw_steps # 3.2 check orphane outputs to see if we can eliminate any required, not_required = scan_utils.scan_can_remove_outs(node.op, orphane_outs) # 3.3. compose replace pairs for those nodes that need not # to store everything in memory ( or ar orphane and required # by the inner function .. ) replaced_outs = [] offset = 1 + op.n_seqs + op.n_mit_mot for idx, _val in enumerate(store_steps[op.n_mit_mot :]): i = idx + op.n_mit_mot if not (type(_val) is int and _val <= 0 and i not in required): if idx + op.n_mit_mot in required: val = 1 else: val = _val # If the memory for this output has been pre-allocated # before going into the scan op (by an alloc node) if idx < op.n_mit_sot + op.n_sit_sot: # In case the input is still an alloc node, we # actually have two options: # a) the input is a set_subtensor, in that case we # can replace the initial tensor by a slice, # b) it is not, and we simply take a slice of it. if nw_inputs[offset + idx].owner and isinstance( nw_inputs[offset + idx].owner.op, tensor.IncSubtensor ): _nw_input = nw_inputs[offset + idx].owner.inputs[1] tmp = pre_greedy_local_optimizer(list_opt_slice, tensor.as_tensor_variable(val - init_l[i])) tmp = pre_constant_merge([tmp])[0] nw_input = scan_utils.expand(_nw_input, tmp) else: tmp = pre_greedy_local_optimizer(list_opt_slice, tensor.as_tensor_variable(val)) tmp = pre_constant_merge([tmp])[0] nw_input = nw_inputs[offset + idx][:tmp] nw_inputs[offset + idx] = nw_input replaced_outs.append(op.n_mit_mot + idx) odx = op.n_mit_mot + idx old_outputs += [(odx, [x[0].outputs[0] for x in node.outputs[odx].clients])] # If there is no memory pre-allocated for this output elif idx < op.n_mit_sot + op.n_sit_sot + op.n_nit_sot: pos = op.n_mit_mot + idx + op.n_seqs + 1 + op.n_shared_outs if nw_inputs[pos] == node.inputs[0]: nw_inputs[pos] = val odx = op.n_mit_mot + idx replaced_outs.append(odx) old_outputs += [(odx, [x[0].outputs[0] for x in node.outputs[odx].clients])] # 3.4. Recompute inputs for everything else based on the new # number of steps if global_nsteps is not None: for idx, val in enumerate(store_steps[op.n_mit_mot :]): if val == 0: if idx < op.n_mit_sot + op.n_sit_sot: _nw_input = nw_inputs[offset + idx].owner.inputs[1] odx = op.n_mit_mot + idx nw_input = scan_utils.expand(_nw_input, nw_steps) nw_inputs[offset + idx] = nw_input elif idx < (op.n_mit_sot + op.n_sit_sot + op.n_nit_sot): in_idx = offset + idx + op.n_shared_outs if nw_inputs[in_idx] == node.inputs[0]: nw_inputs[in_idx] = nw_steps odx = op.n_mit_mot + idx # 3.5 Remove unwanted orphane outputs (inps, outs, info, node_ins, compress_map) = scan_utils.compress_outs(op, not_required, nw_inputs) inv_compress_map = {} for k, v in compress_map.items(): inv_compress_map[v] = k node_ins = [pre_greedy_local_optimizer(list_opt_slice, x) for x in node_ins] node_ins = pre_constant_merge(node_ins) # 3.6 Compose the new scan # I need to make sure I'm not reapplying the same optimization # twice since bad things usually happen if I do that info["_scan_merge_visited"] = True new_outs = scan_op.Scan(inps, outs, info).make_node(*node_ins).outputs old_new = [] # 3.7 Get replace pairs for those outputs that do not change # the number of intermediate steps stored for idx, sl in enumerate(slices): if global_nsteps and sl is not None and store_steps[idx] == 0: for hdx, cl in enumerate(node.outputs[idx].clients): cnf_slice, old_slices = sl[hdx] # Sanitize the nw_slice by converting ints back into # constants :) I only need to do this for the first # slice since that is the only slice if isinstance(cnf_slice[0], slice): fslice = slice( sanitize(cnf_slice[0].start), sanitize(cnf_slice[0].stop), sanitize(cnf_slice[0].step) ) else: fslice = sanitize(cnf_slice[0]) nw_slice = (fslice,) + tuple(old_slices[1:]) nw_pos = inv_compress_map[idx] nw_out = new_outs[nw_pos] subtens = tensor.basic.Subtensor(nw_slice) # slice inputs sl_ins = tensor.basic.Subtensor.collapse( nw_slice, lambda entry: isinstance(entry, tensor.Variable) ) new_o = subtens.make_node(new_outs[nw_pos], *sl_ins).outputs[0] if new_o.ndim > 0: new_o = new_o[:: cnf_slice[1]] replaced_outs.append(idx) old_new += [(cl[0].outputs[0], new_o)] # 3.8. Get replace pairs for those outputs that change # the number of stored intermediate steps for pos, old_outs in old_outputs: if len(old_outs) > 0: nw_pos = compress_map[pos] nw_out = new_outs[nw_pos] for k, old in enumerate(old_outs): # Get the correct slice cnf_slice, old_slices = slices[pos][k] if type(cnf_slice[0]) is slice: start = cnf_slice[0].start - nw_steps - init_l[pos] + store_steps[pos] if cnf_slice[0].stop is not None and cnf_slice[0].stop != sys.maxint: stop = cnf_slice[0].stop - nw_steps - init_l[pos] + store_steps[pos] else: stop = None nw_slice = (slice(sanitize(start), sanitize(stop), sanitize(cnf_slice[0].step)),) + tuple( old_slices[1:] ) else: position = cnf_slice[0] - nw_steps - init_l[pos] + store_steps[pos] nw_slice = (sanitize(position),) + tuple(old_slices[1:]) subtens = tensor.basic.Subtensor(nw_slice) sl_ins = tensor.basic.Subtensor.collapse( nw_slice, lambda entry: isinstance(entry, tensor.Variable) ) new_o = subtens.make_node(new_outs[nw_pos], *sl_ins).outputs[0] if new_o.ndim > 0: new_o = new_o[:: cnf_slice[1]] old_new += [(old, new_o)] # 3.9. Get replace pairs for all other nodes if flag_store or global_nsteps is not None: for idx, o in enumerate(node.outputs): if not (idx in replaced_outs) and not idx in not_required: nw_pos = compress_map[idx] old_new += [(o, new_outs[nw_pos])] env.replace_all_validate(old_new, reason="scan_save_mem")
# No need to print a warning or raise an error now, # it will be done when fn will be called. _logger.info(('Cannot compute test value for the ' 'inner function of scan, input value missing %s'), e) if getattr(init_out['initial'], 'name', None) is not None: arg.name = init_out['initial'].name + '[t-1]' # We need now to allocate space for storing the output and copy # the initial state over. We do this using the expand function # defined in scan utils sit_sot_scan_inputs.append( scan_utils.expand( tensor.unbroadcast( tensor.shape_padleft(actual_arg), 0), actual_n_steps )) sit_sot_inner_slices.append(actual_arg) if i in return_steps: sit_sot_return_steps[n_sit_sot] = return_steps[i] sit_sot_inner_inputs.append(arg) sit_sot_rightOrder.append(i) n_sit_sot += 1 elif init_out.get('taps', None): if numpy.any(numpy.array(init_out.get('taps', [])) > 0): # Make sure we do not have requests for future values of a # sequence we can not provide such values
def process_node(self, fgraph, node): # helpful functions def select_min(x, y): if x is None: return y if y is None: return x return tensor.minimum(x, y) def select_max(x, y): if x is None: return y if y is None: return x return tensor.maximum(x, y) def sanitize(x): if x is None: return None else: return tensor.as_tensor_variable(x) if hasattr(fgraph, 'shape_feature'): shape_of = node.fgraph.shape_feature.shape_of else: # Each access to shape_of is in a try..except block in order to # use a default version when the variable is not in the shape_of # dictionary. shape_of = {} # 1. Initialization of variables # Note 1) We do not actually care about outputs representing shared # variables (those have no intermediate values) so it is safer to # ignore them and not change them in any way. To simplify the # optimizations I construct the variable ``c_outs`` ( that counts # outputs up to those we care) and the list ``init_l`` which for any # output we care says the length of its initial state. Note that # defining ``init_l`` for mit_mot sequences is a bit trickier but # it is safe to set it to 0 op = node.op c_outs = op.n_mit_mot + op.n_mit_sot + op.n_sit_sot + op.n_nit_sot init_l = [0 for x in xrange(op.n_mit_mot)] init_l += [abs(numpy.min(v)) for v in op.tap_array[op.n_mit_mot:]] init_l += [0 for x in xrange(op.n_nit_sot)] # 2. Check the clients of each output and see for how many steps # does scan need to run # This comparison checks if there is any uncounted output, which # can only be an output corresponding to a shared variable # 2.1 Initialize # global_nsteps is a dictionary having two fields ( 'real' deals # with int values, 'sym' with symbolic ones) or None # given that a scan op has k outputs o_1, .. o_k and each # output has n_j clients c_1^1, c_1^2, .. c_1^{n_1}, c_2^1, .., # global_nsteps is None if any of the clients is different # from a subtensor or its real and sym field equal to # max(c_i_j.idx_list[0].stop), meaning store up to which maximal # index(step) for any output scan actually needs to compute # In other words n_steps should be equal to this maximal ! # Note: if we have a shared variable that gets updated at every step # of the loop, reducing the number of steps will affect the the # value of the shared variable after the loop so we need not to # change the number of steps in that case. To do this we set # global_nsteps to None which is seen as a flag that nothing needs # to be done assert len(node.outputs) >= c_outs if len(node.outputs) == c_outs: global_nsteps = {'real': -1, 'sym': []} else: global_nsteps = None # Keeps track of the original slices that each client represent slices = [None for o in node.outputs] # A list for each output indicating how many intermediate values # should be stored. If negative it means none of the intermediate # values (i.e. the output can be removed since it is not used # afterwards in the computations), if 0 it means that all # intermediate values are required, otherwise is up to that number # of intermediate values # Note that for mit_mot outputs and shared outputs we can not change # the number of intermediate steps stored without affecting the # result of the op store_steps = [0 for o in xrange(op.n_mit_mot)] store_steps += [-1 for o in node.outputs[op.n_mit_mot:c_outs]] # Flag that says if an input has changed and we need to do something # or not flag_store = False # 2.2 Loop over the clients for i, out in enumerate(node.outputs[:c_outs]): # look at all its clients slices[i] = [] for cl, _ in out.clients: # 2.1 outputs of the function #=> output needs all its intermediate values if type(cl) == str: # if the node is actually an output, then # we need to store the entire thing global_nsteps = None slices[i] = None break # 2.2 non-subtensor nodes #=> output needs all its intermediate values elif not isinstance(cl.op, tensor.basic.Subtensor): global_nsteps = None slices[i] = None break # 2.3 subtensor nodes #=> output might need to store just a subset of its values else: # 2.3.1 extract idx list of subtensor this_slice = tensor.basic.get_idx_list(cl.inputs, cl.op.idx_list) if this_slice is None: # if unable to extract idx_list #=> outputs needs all its intermediate values global_nsteps = None slices[i] = None break # 2.3.2 extract the begin/end of the first dimension if i >= op.n_mit_mot: try: length = shape_of[out][0] except KeyError: length = node.inputs[0] + init_l[i] else: try: length = shape_of[out][0] except KeyError: length = out.shape[0] cf_slice = tensor.basic.get_canonical_form_slice( this_slice[0], length) slices[i] += [(cf_slice, this_slice)] if (isinstance(this_slice[0], slice) and this_slice[0].stop is None): global_nsteps = None if isinstance(cf_slice[0], slice): stop = tensor.basic.extract_constant(cf_slice[0].stop) else: stop = tensor.basic.extract_constant(cf_slice[0]) + 1 if stop == maxsize or stop == length: stop = None else: # there is a **gotcha** here ! Namely, scan returns an # array that contains the initial state of the output # as well. Which means that if have a initial state of # length 3, and you look for 5 steps you get an output # y of length 8. If you only use y[:5], this does not # mean that you only need to loop for 5 steps but # actually only for 2 steps ( the first 3 are the # initial state) stop = stop - init_l[i] # 2.3.3 we might get away with less number of steps if stop is not None and global_nsteps is not None: # yes if it is a tensor if isinstance(stop, tensor.Variable): global_nsteps['sym'] += [stop] # not if it is maxsize elif (type(stop) in (int, long) and stop == maxsize): global_nsteps = None # yes if it is a int k, 0 < k < maxsize elif (type(stop) in (int, long) and global_nsteps['real'] < stop): global_nsteps['real'] = stop # yes if it is a int k, 0 < k < maxsize elif (type(stop) in (int, long) and stop > 0): pass # not otherwise else: global_nsteps = None # 2.3. Analyze global_nsteps to figure out for how many steps scan # needs to iterate if global_nsteps is not None: nw_steps = node.inputs[0] # there are some symbolic tensors that limit the number of # steps if len(global_nsteps['sym']) == 0: sym_steps = None else: sym_steps = global_nsteps['sym'][0] for c in global_nsteps['sym'][1:]: sym_steps = tensor.maximum(sym_steps, c) if global_nsteps['real'] >= 0: real_steps = global_nsteps['real'] else: real_steps = None nw_steps = select_min(select_max(sym_steps, real_steps), node.inputs[0]) else: nw_steps = node.inputs[0] global_nsteps = None # 2.4 Loop over the clients again now looking just to see how many # intermediate steps to store for i, out in enumerate(node.outputs[:c_outs]): # look at all its clients for cl, _ in out.clients: if type(cl) == str: store_steps[i] = 0 break elif not isinstance(cl.op, tensor.basic.Subtensor): store_steps[i] = 0 break else: this_slice = tensor.basic.get_idx_list(cl.inputs, cl.op.idx_list) if this_slice is None: store_steps[i] = 0 break if (isinstance(this_slice[0], slice) and this_slice[0].start is None): store_steps[i] = 0 break if i > op.n_mit_mot: length = node.inputs[0] + init_l[i] else: try: length = shape_of[out][0] except KeyError: length = out.shape[0] cf_slice = tensor.basic.get_canonical_form_slice( this_slice[0], length) if isinstance(cf_slice[0], slice): start = tensor.basic.extract_constant( cf_slice[0].start) else: start = tensor.basic.extract_constant(cf_slice[0]) if start == 0 or store_steps[i] == 0: store_steps[i] = 0 else: pval = select_max(nw_steps - start + init_l[i], init_l[i]) if store_steps[i] != -1: pval = select_max(pval, store_steps[i]) store_steps[i] = pval flag_store = True orphane_outs = [i for i, x in enumerate(store_steps) if (type(x) is int) and (x < 0)] flag_store = flag_store or (len(orphane_outs) > 0) # 3. is there anything to change ? if (flag_store or global_nsteps is not None): # 3.1 initialize inputs for the new scan old_outputs = [] nw_inputs = list(node.inputs) nw_inputs[0] = nw_steps # 3.2 check orphane outputs to see if we can eliminate any required, not_required = \ scan_utils.scan_can_remove_outs(node.op, orphane_outs) # 3.3. compose replace pairs for those nodes that need not # to store everything in memory ( or ar orphane and required # by the inner function .. ) replaced_outs = [] offset = 1 + op.n_seqs + op.n_mit_mot for idx, _val in enumerate(store_steps[op.n_mit_mot:]): i = idx + op.n_mit_mot if not(type(_val) is int and _val <= 0 and i not in required): if idx + op.n_mit_mot in required: val = 1 else: val = _val # If the memory for this output has been pre-allocated # before going into the scan op (by an alloc node) if idx < op.n_mit_sot + op.n_sit_sot: # In case the input is still an alloc node, we # actually have two options: # a) the input is a set_subtensor, in that case we # can replace the initial tensor by a slice, # b) it is not, and we simply take a slice of it. #TODO: commit change below with Razvan if (nw_inputs[offset + idx].owner and isinstance(nw_inputs[offset + idx].owner.op, tensor.IncSubtensor) and isinstance(nw_inputs[offset+idx].owner.op.idx_list[0], slice)): _nw_input = nw_inputs[offset + idx].owner.inputs[1] val = tensor.as_tensor_variable(val) initl = tensor.as_tensor_variable(init_l[i]) tmp = pre_greedy_local_optimizer(list_opt_slice, tensor.maximum(val - initl, 0)) tmp = pre_constant_merge([tmp])[0] nw_input = scan_utils.expand(_nw_input, tmp) else: tmp = tensor.as_tensor_variable(val) initl = tensor.as_tensor_variable(init_l[i]) tmp = tensor.maximum(tmp, initl) tmp = pre_greedy_local_optimizer(list_opt_slice, tmp) tmp = pre_constant_merge([tmp])[0] nw_input = nw_inputs[offset + idx][:tmp] nw_inputs[offset + idx] = nw_input replaced_outs.append(op.n_mit_mot + idx) odx = op.n_mit_mot + idx old_outputs += [(odx, [x[0].outputs[0] for x in node.outputs[odx].clients])] # If there is no memory pre-allocated for this output elif idx < op.n_mit_sot + op.n_sit_sot + op.n_nit_sot: pos = (op.n_mit_mot + idx + op.n_seqs + 1 + op.n_shared_outs) if nw_inputs[pos] == node.inputs[0]: nw_inputs[pos] = val odx = op.n_mit_mot + idx replaced_outs.append(odx) old_outputs += [(odx, [x[0].outputs[0] for x in node.outputs[odx].clients])] # 3.4. Recompute inputs for everything else based on the new # number of steps if global_nsteps is not None: for idx, val in enumerate(store_steps[op.n_mit_mot:]): if val == 0: if idx < op.n_mit_sot + op.n_sit_sot: _nw_input = nw_inputs[offset + idx].owner.inputs[1] odx = op.n_mit_mot + idx nw_input = scan_utils.expand(_nw_input, nw_steps) nw_inputs[offset + idx] = nw_input elif idx < (op.n_mit_sot + op.n_sit_sot + op.n_nit_sot): in_idx = offset + idx + op.n_shared_outs if nw_inputs[in_idx] == node.inputs[0]: nw_inputs[in_idx] = nw_steps odx = op.n_mit_mot + idx # 3.5 Remove unwanted orphane outputs (inps, outs, info, node_ins, compress_map) = \ scan_utils.compress_outs(op, not_required, nw_inputs) inv_compress_map = {} for k, v in compress_map.items(): inv_compress_map[v] = k node_ins = [pre_greedy_local_optimizer(list_opt_slice, x) for x in node_ins] node_ins = pre_constant_merge(node_ins) # 3.6 Compose the new scan # I need to make sure I'm not reapplying the same optimization # twice since bad things usually happen if I do that info['_scan_savemem_visited'] = True new_outs = scan_op.Scan(inps, outs, info).make_node(*node_ins).outputs old_new = [] # 3.7 Get replace pairs for those outputs that do not change # the number of intermediate steps stored for idx, sl in enumerate(slices): if global_nsteps and sl is not None and store_steps[idx] == 0: for hdx, cl in enumerate(node.outputs[idx].clients): cnf_slice, old_slices = sl[hdx] # Sanitize the nw_slice by converting ints back into # constants :) I only need to do this for the first # slice since that is the only slice if isinstance(cnf_slice[0], slice): fslice = slice( sanitize(cnf_slice[0].start), sanitize(cnf_slice[0].stop), sanitize(cnf_slice[0].step)) else: fslice = sanitize(cnf_slice[0]) nw_slice = (fslice,) + tuple(old_slices[1:]) nw_pos = inv_compress_map[idx] subtens = tensor.basic.Subtensor(nw_slice) # slice inputs sl_ins = tensor.basic.Subtensor.collapse( nw_slice, lambda entry: isinstance(entry, tensor.Variable)) new_o = subtens.make_node(new_outs[nw_pos], *sl_ins).outputs[0] if new_o.ndim > 0: new_o = new_o[::cnf_slice[1]] replaced_outs.append(idx) old_new += [(cl[0].outputs[0], new_o)] # 3.8. Get replace pairs for those outputs that change # the number of stored intermediate steps for pos, old_outs in old_outputs: if len(old_outs) > 0: nw_pos = compress_map[pos] for k, old in enumerate(old_outs): # Get the correct slice cnf_slice, old_slices = slices[pos][k] if type(cnf_slice[0]) is slice: start = (cnf_slice[0].start - nw_steps - init_l[pos] + store_steps[pos]) if (cnf_slice[0].stop is not None and cnf_slice[0].stop != maxsize): stop = (cnf_slice[0].stop - nw_steps - init_l[pos] + store_steps[pos]) else: stop = None nw_slice = ((slice(sanitize(start), sanitize(stop), sanitize(cnf_slice[0].step)),) + tuple(old_slices[1:])) else: position = (cnf_slice[0] - nw_steps - init_l[pos] + store_steps[pos]) nw_slice = (sanitize(position),) + \ tuple(old_slices[1:]) subtens = tensor.basic.Subtensor(nw_slice) sl_ins = tensor.basic.Subtensor.collapse( nw_slice, lambda entry: isinstance(entry, tensor.Variable)) new_o = subtens.make_node(new_outs[nw_pos], *sl_ins).outputs[0] if new_o.ndim > 0: new_o = new_o[::cnf_slice[1]] old_new += [(old, new_o)] # 3.9. Get replace pairs for all other nodes if flag_store or global_nsteps is not None: for idx, o in enumerate(node.outputs): if not (idx in replaced_outs) and not idx in not_required: nw_pos = compress_map[idx] old_new += [(o, new_outs[nw_pos])] fgraph.replace_all_validate_remove(old_new, remove=[node], reason='scan_save_mem')
def scan(fn, sequences=None, outputs_info=None, non_sequences=None, n_steps=None, truncate_gradient=-1, go_backwards=False, mode=None, name=None, options=None, profile=False): """ This function constructs and applies a Scan op to the provided arguments. :param fn: ``fn`` is a function that describes the operations involved in one step of ``scan``. ``fn`` should construct variables describing the output of one iteration step. It should expect as input theano variables representing all the slices of the input sequences and previous values of the outputs, as well as all other arguments given to scan as ``non_sequences``. The order in which scan passes these variables to ``fn`` is the following : * all time slices of the first sequence * all time slices of the second sequence * ... * all time slices of the last sequence * all past slices of the first output * all past slices of the second otuput * ... * all past slices of the last output * all other arguments (the list given as `non_sequences` to scan) The order of the sequences is the same as the one in the list `sequences` given to scan. The order of the outputs is the same as the order of ``output_info``. For any sequence or output the order of the time slices is the same as the one in which they have been given as taps. For example if one writes the following : .. code-block:: python scan(fn, sequences = [ dict(input= Sequence1, taps = [-3,2,-1]) , Sequence2 , dict(input = Sequence3, taps = 3) ] , outputs_info = [ dict(initial = Output1, taps = [-3,-5]) , dict(initial = Output2, taps = None) , Output3 ] , non_sequences = [ Argument1, Argument 2]) ``fn`` should expect the following arguments in this given order: #. ``Sequence1[t-3]`` #. ``Sequence1[t+2]`` #. ``Sequence1[t-1]`` #. ``Sequence2[t]`` #. ``Sequence3[t+3]`` #. ``Output1[t-3]`` #. ``Output1[t-5]`` #. ``Output3[t-1]`` #. ``Argument1`` #. ``Argument2`` The list of ``non_sequences`` can also contain shared variables used in the function, though ``scan`` is able to figure those out on its own so they can be skipped. For the clarity of the code we recommand though to provide them to scan. To some extend ``scan`` can also figure out other ``non sequences`` (not shared) even if not passed to scan (but used by `fn`). A simple example of this would be : .. code-block:: python import theano.tensor as TT W = TT.matrix() W_2 = W**2 def f(x): return TT.dot(x,W_2) The function is expected to return two things. One is a list of outputs ordered in the same order as ``outputs_info``, with the difference that there should be only one output variable per output initial state (even if no tap value is used). Secondly `fn` should return an update dictionary (that tells how to update any shared variable after each iteration step). The dictionary can optionally be given as a list of tuples. There is no constraint on the order of these two list, ``fn`` can return either ``(outputs_list, update_dictionary)`` or ``(update_dictionary, outputs_list)`` or just one of the two (in case the other is empty). To use ``scan`` as a while loop, the user needs to change the function ``fn`` such that also a stopping condition is returned. To do so, he/she needs to wrap the condition in an ``until`` class. The condition should be returned as a third element, for example: .. code-block:: python ... return [y1_t, y2_t], {x:x+1}, theano.scan_module.until(x < 50) Note that a number of steps (considered in here as the maximum number of steps ) is still required even though a condition is passed (and it is used to allocate memory if needed). = {}): :param sequences: ``sequences`` is the list of Theano variables or dictionaries describing the sequences ``scan`` has to iterate over. If a sequence is given as wrapped in a dictionary, then a set of optional information can be provided about the sequence. The dictionary should have the following keys: * ``input`` (*mandatory*) -- Theano variable representing the sequence. * ``taps`` -- Temporal taps of the sequence required by ``fn``. They are provided as a list of integers, where a value ``k`` impiles that at iteration step ``t`` scan will pass to ``fn`` the slice ``t+k``. Default value is ``[0]`` Any Theano variable in the list ``sequences`` is automatically wrapped into a dictionary where ``taps`` is set to ``[0]`` :param outputs_info: ``outputs_info`` is the list of Theano variables or dictionaries describing the initial state of the outputs computed recurrently. When this initial states are given as dictionary optional information can be provided about the output corresponding to these initial states. The dictionary should have the following keys: * ``initial`` -- Theano variable that represents the initial state of a given output. In case the output is not computed recursively (think of a map) and does not require a initial state this field can be skiped. Given that only the previous time step of the output is used by ``fn`` the initial state should have the same shape as the output. If multiple time taps are used, the initial state should have one extra dimension that should cover all the possible taps. For example if we use ``-5``, ``-2`` and ``-1`` as past taps, at step 0, ``fn`` will require (by an abuse of notation) ``output[-5]``, ``output[-2]`` and ``output[-1]``. This will be given by the initial state, which in this case should have the shape (5,)+output.shape. If this variable containing the initial state is called ``init_y`` then ``init_y[0]`` *corresponds to* ``output[-5]``. ``init_y[1]`` *correponds to* ``output[-4]``, ``init_y[2]`` corresponds to ``output[-3]``, ``init_y[3]`` coresponds to ``output[-2]``, ``init_y[4]`` corresponds to ``output[-1]``. While this order might seem strange, it comes natural from splitting an array at a given point. Assume that we have a array ``x``, and we choose ``k`` to be time step ``0``. Then our initial state would be ``x[:k]``, while the output will be ``x[k:]``. Looking at this split, elements in ``x[:k]`` are ordered exactly like those in ``init_y``. * ``taps`` -- Temporal taps of the output that will be pass to ``fn``. They are provided as a list of *negative* integers, where a value ``k`` implies that at iteration step ``t`` scan will pass to ``fn`` the slice ``t+k``. ``scan`` will follow this logic if partial information is given: * If an output is not wrapped in a dictionary, ``scan`` will wrap it in one assuming that you use only the last step of the output (i.e. it makes your tap value list equal to [-1]). * If you wrap an output in a dictionary and you do not provide any taps but you provide an initial state it will assume that you are using only a tap value of -1. * If you wrap an output in a dictionary but you do not provide any initial state, it assumes that you are not using any form of taps. * If you provide a ``None`` instead of a variable or a empty dictionary ``scan`` assumes that you will not use any taps for this output (like for example in case of a map) If ``outputs_info`` is an empty list or None, ``scan`` assumes that no tap is used for any of the outputs. If information is provided just for a subset of the outputs an exception is raised (because there is no convention on how scan should map the provided information to the outputs of ``fn``) :param non_sequences: ``non_sequences`` is the list of arguments that are passed to ``fn`` at each steps. One can opt to exclude variable used in ``fn`` from this list as long as they are part of the computational graph, though for clarity we encourage not to do so. :param n_steps: ``n_steps`` is the number of steps to iterate given as an int or Theano scalar. If any of the input sequences do not have enough elements, scan will raise an error. If the *value is 0* the outputs will have *0 rows*. If the value is negative, ``scan`` will run backwards in time. If the ``go_backwards`` flag is already set and also ``n_steps`` is negative, ``scan`` will run forward in time. If n stpes is not provided, ``scan`` will figure out the amount of steps it should run given its input sequences. :param truncate_gradient: ``truncate_gradient`` is the number of steps to use in truncated BPTT. If you compute gradients through a scan op, they are computed using backpropagation through time. By providing a different value then -1, you choose to use truncated BPTT instead of classical BPTT, where you go for only ``truncate_gradient`` number of steps back in time. :param go_backwards: ``go_backwards`` is a flag indicating if ``scan`` should go backwards through the sequences. If you think of each sequence as indexed by time, making this flag True would mean that ``scan`` goes back in time, namely that for any sequence it starts from the end and goes towards 0. :param name: When profiling ``scan``, it is crucial to provide a name for any instance of ``scan``. The profiler will produce an overall profile of your code as well as profiles for the computation of one step of each instance of ``scan``. The ``name`` of the instance appears in those profiles and can greatly help to disambiguate information. :param mode: It is recommended to leave this argument to None, especially when profiling ``scan`` (otherwise the results are not going to be accurate). If you prefer the computations of one step of ``scan`` to be done differently then the entire function, you can use this parameter to describe how the computations in this loop are done (see ``theano.function`` for details about possible values and their meaning). :param profile: Flag or string. If true, or different from the empty string, a profile object will be created and attached to the inner graph of scan. In case ``profile`` is True, the profile object will have the name of the scan instance, otherwise it will have the passed string. Profile object collect (and print) information only when running the inner graph with the new cvm linker ( with default modes, other linkers this argument is useless) :rtype: tuple :return: tuple of the form (outputs, updates); ``outputs`` is either a Theano variable or a list of Theano variables representing the outputs of ``scan`` (in the same order as in ``outputs_info``). ``updates`` is a subclass of dictionary specifying the update rules for all shared variables used in scan This dictionary should be passed to ``theano.function`` when you compile your function. The change compared to a normal dictionary is that we validate that keys are SharedVariable and addition of those dictionary are validated to be consistent. """ # Note : see the internal documentation of the scan op for naming # conventions and all other details if options is None: options = {} rvals = scan_utils.canonical_arguments(sequences, outputs_info, non_sequences, go_backwards, n_steps) inputs, states_and_outputs_info, parameters, T = rvals # If we provided a known number of steps ( before compilation) # and if that number is 1 or -1, then we can skip the Scan Op, # and just apply the inner function once # To do that we check here to see the nature of n_steps T_value = None if isinstance(n_steps, (float, int)): T_value = int(n_steps) else: try: T_value = opt.get_constant_value(n_steps) except (TypeError, AttributeError): T_value = None if T_value in (1, -1): return one_step_scan(fn, inputs, states_and_outputs_info, parameters, truncate_gradient) # 1. Variable representing the current time step t = scalar_shared(numpy.int64(0), name='t') # 2. Allocate memory for the states of scan. mintaps = [] lengths = [] for pos, arg_info in enumerate(states_and_outputs_info): if arg_info.get('taps', None) == [-1]: mintaps.append(1) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) arg_info['initial'] = scan_utils.expand(tensor.unbroadcast( tensor.shape_padleft(arg_info['initial']), 0), T) elif arg_info.get('taps', None): if numpy.any(numpy.array(arg_info.get('taps', [])) > 0): # Make sure we do not have requests for future values of a # sequence we can not provide such values raise ValueError('Can not use future taps of outputs', arg_info) mintap = abs(numpy.min(arg_info['taps'])) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) mintaps.append(mintap) arg_info['initial'] = scan_utils.expand( arg_info['initial'][:mintap], T) else: mintaps.append(0) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) # 3. Generate arguments for the function passed to scan. This will # function will return the outputs that need to be computed at every # timesteps inputs_slices = [input[t] for input in inputs] states_slices = [] for n, state in enumerate(states_and_outputs_info): # Check if it is actually a state and not an output if mintaps[n] != 0: for k in state['taps']: states_slices.append( state['initial'][(t + mintaps[n] + k) % lengths[n]]) # 4. Construct outputs that are to be computed by the inner # function of scan args = inputs_slices + states_slices + parameters cond, states_and_outputs, updates = \ scan_utils.get_updates_and_outputs(fn(*args)) # User is allowed to provide no information if it only behaves like a # map if (len(states_and_outputs) != len(states_and_outputs_info) and len(states_and_outputs_info) == 0): mintaps = [0] * len(states_and_outputs) # 5. Construct the scan op # 5.1 Construct list of shared variables with updates (those that # can be treated as states (i.e. of TensorType) and those that can not # (like Random States) if cond is not None: _cond = [cond] else: _cond = [] rvals = rebuild_collect_shared( states_and_outputs + _cond, updates=updates, rebuild_strict=True, copy_inputs_over=True, no_default_updates=False) # extracting the arguments input_variables, cloned_outputs, other_rval = rvals clone_d, update_d, update_expr, shared_inputs = other_rval additional_input_states = [] additional_output_states = [] additional_lengths = [] additional_mintaps = [] original_numeric_shared_variables = [] non_numeric_input_states = [] non_numeric_output_states = [] original_non_numeric_shared_variables = [] pos = len(lengths) for sv in shared_inputs: if sv in update_d: if isinstance(sv, (TensorVariable, TensorSharedVariable)): # We can treat it as a sit sot nw_state = scan_utils.expand( tensor.unbroadcast(tensor.shape_padleft(sv), 0), T) additional_lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) pos = pos + 1 additional_mintaps.append(1) additional_input_states.append(nw_state) additional_output_states.append( scan_utils.clone(tensor.set_subtensor( nw_state[(t + 1) % additional_lengths[-1]], update_d[sv]))) original_numeric_shared_variables.append(sv) else: non_numeric_input_states.append(sv) non_numeric_output_states.append(update_d[sv]) original_non_numeric_shared_variables.append(sv) # Replace shared variables in the update _additional_output_states = [] replace = {} for sv, buf in zip(original_numeric_shared_variables, additional_input_states): replace[sv] = buf[t] for out in additional_output_states: _additional_output_states.append( scan_utils.clone(out, replace=replace)) additional_output_states = _additional_output_states # 5.2 Collect inputs/outputs of the inner function inputs = [] outputs = [] for n, mintap in enumerate(mintaps): if mintap != 0: input_state = states_and_outputs_info[n]['initial'] inputs.append(input_state) outputs.append( tensor.set_subtensor( input_state[(t + mintap) % lengths[n]], states_and_outputs[n])) else: mem_buffer = scan_utils.allocate_memory( T, states_and_outputs_info[n], states_and_outputs[n]) inputs.append(output) outputs.append( tensor.set_subtensor(output[t % lengths[n]], states_and_outputs[n])) inputs.extend(additional_input_states) outputs.extend(additional_output_states) lengths.extend(additional_lengths) mintaps.extend(additional_mintaps) inputs.extend(non_numeric_input_states) outputs.extend(non_numeric_output_states) all_other_inputs = gof.graph.inputs(outputs) parameters = [x for x in all_other_inputs if (x not in inputs and x not in lengths and x is not t and isinstance(x, gof.Variable) and not isinstance(x, gof.Constant))] inputs.extend(parameters) # 5.3 Construct the the options dictionary options['name'] = name options['profile'] = profile options['mode'] = mode options['inplace'] = False options['gpu'] = False options['truncate_gradient'] = truncate_gradient options['hash_inner_graph'] = 0 # 5.4 Construct the ScanOp instance local_op = scan_op.ScanOp(inputs=inputs, outputs=outputs, lengths=lengths, switches=[], mintaps=mintaps, index=t, options=options, as_repeatUntil=cond) # Note that we get here all the outputs followed by the update rules to # the shared variables we had in our scan # we know that we have (in this given order): # * len(states_and_outputs) real outputs # * len(additional_input_states) updates for numeric shared variable # * len(non_numeric_input_states) updates for non numeric shared # variables scan_inputs = [T] + inputs scan_outputs_update_rules = scan_utils.to_list(local_op(*scan_inputs)) # 5.5 Collect outputs and add permutation object scan_outputs = [] for pos in xrange(len(states_and_outputs)): out = scan_utils.ScanPermutation(mintaps[pos])( scan_outputs_update_rules[pos], t) scan_outputs.append(out[mintaps[pos]:]) # 5.6 Construct updates dictionary update_rules = scan_outputs_update_rules[len(states_and_outputs):] updates = {} for v, u in izip(original_numeric_shared_variables, update_rules[:len(additional_input_states)]): updates[v] = u[-1] for v, u in izip(original_non_numeric_shared_variables, update_rules[len(additional_input_states):]): updates[v] = u # Step 5.7 We are done and can return everything back to the user return scan_outputs, updates
def scan(fn, sequences=None, outputs_info=None, non_sequences=None, n_steps=None, truncate_gradient=-1, go_backwards=False, mode=None, name=None, options=None, profile=False): """ This function constructs and applies a Scan op to the provided arguments. :param fn: ``fn`` is a function that describes the operations involved in one step of ``scan``. ``fn`` should construct variables describing the output of one iteration step. It should expect as input theano variables representing all the slices of the input sequences and previous values of the outputs, as well as all other arguments given to scan as ``non_sequences``. The order in which scan passes these variables to ``fn`` is the following : * all time slices of the first sequence * all time slices of the second sequence * ... * all time slices of the last sequence * all past slices of the first output * all past slices of the second otuput * ... * all past slices of the last output * all other arguments (the list given as `non_sequences` to scan) The order of the sequences is the same as the one in the list `sequences` given to scan. The order of the outputs is the same as the order of ``outputs_info``. For any sequence or output the order of the time slices is the same as the one in which they have been given as taps. For example if one writes the following : .. code-block:: python scan(fn, sequences = [ dict(input= Sequence1, taps = [-3,2,-1]) , Sequence2 , dict(input = Sequence3, taps = 3) ] , outputs_info = [ dict(initial = Output1, taps = [-3,-5]) , dict(initial = Output2, taps = None) , Output3 ] , non_sequences = [ Argument1, Argument 2]) ``fn`` should expect the following arguments in this given order: #. ``Sequence1[t-3]`` #. ``Sequence1[t+2]`` #. ``Sequence1[t-1]`` #. ``Sequence2[t]`` #. ``Sequence3[t+3]`` #. ``Output1[t-3]`` #. ``Output1[t-5]`` #. ``Output3[t-1]`` #. ``Argument1`` #. ``Argument2`` The list of ``non_sequences`` can also contain shared variables used in the function, though ``scan`` is able to figure those out on its own so they can be skipped. For the clarity of the code we recommend though to provide them to scan. To some extend ``scan`` can also figure out other ``non sequences`` (not shared) even if not passed to scan (but used by `fn`). A simple example of this would be : .. code-block:: python import theano.tensor as TT W = TT.matrix() W_2 = W**2 def f(x): return TT.dot(x,W_2) The function is expected to return two things. One is a list of outputs ordered in the same order as ``outputs_info``, with the difference that there should be only one output variable per output initial state (even if no tap value is used). Secondly `fn` should return an update dictionary (that tells how to update any shared variable after each iteration step). The dictionary can optionally be given as a list of tuples. There is no constraint on the order of these two list, ``fn`` can return either ``(outputs_list, update_dictionary)`` or ``(update_dictionary, outputs_list)`` or just one of the two (in case the other is empty). To use ``scan`` as a while loop, the user needs to change the function ``fn`` such that also a stopping condition is returned. To do so, he/she needs to wrap the condition in an ``until`` class. The condition should be returned as a third element, for example: .. code-block:: python ... return [y1_t, y2_t], {x:x+1}, theano.scan_module.until(x < 50) Note that a number of steps (considered in here as the maximum number of steps ) is still required even though a condition is passed (and it is used to allocate memory if needed). = {}): :param sequences: ``sequences`` is the list of Theano variables or dictionaries describing the sequences ``scan`` has to iterate over. If a sequence is given as wrapped in a dictionary, then a set of optional information can be provided about the sequence. The dictionary should have the following keys: * ``input`` (*mandatory*) -- Theano variable representing the sequence. * ``taps`` -- Temporal taps of the sequence required by ``fn``. They are provided as a list of integers, where a value ``k`` impiles that at iteration step ``t`` scan will pass to ``fn`` the slice ``t+k``. Default value is ``[0]`` Any Theano variable in the list ``sequences`` is automatically wrapped into a dictionary where ``taps`` is set to ``[0]`` :param outputs_info: ``outputs_info`` is the list of Theano variables or dictionaries describing the initial state of the outputs computed recurrently. When this initial states are given as dictionary optional information can be provided about the output corresponding to these initial states. The dictionary should have the following keys: * ``initial`` -- Theano variable that represents the initial state of a given output. In case the output is not computed recursively (think of a map) and does not require a initial state this field can be skiped. Given that only the previous time step of the output is used by ``fn`` the initial state should have the same shape as the output. If multiple time taps are used, the initial state should have one extra dimension that should cover all the possible taps. For example if we use ``-5``, ``-2`` and ``-1`` as past taps, at step 0, ``fn`` will require (by an abuse of notation) ``output[-5]``, ``output[-2]`` and ``output[-1]``. This will be given by the initial state, which in this case should have the shape (5,)+output.shape. If this variable containing the initial state is called ``init_y`` then ``init_y[0]`` *corresponds to* ``output[-5]``. ``init_y[1]`` *correponds to* ``output[-4]``, ``init_y[2]`` corresponds to ``output[-3]``, ``init_y[3]`` coresponds to ``output[-2]``, ``init_y[4]`` corresponds to ``output[-1]``. While this order might seem strange, it comes natural from splitting an array at a given point. Assume that we have a array ``x``, and we choose ``k`` to be time step ``0``. Then our initial state would be ``x[:k]``, while the output will be ``x[k:]``. Looking at this split, elements in ``x[:k]`` are ordered exactly like those in ``init_y``. * ``taps`` -- Temporal taps of the output that will be pass to ``fn``. They are provided as a list of *negative* integers, where a value ``k`` implies that at iteration step ``t`` scan will pass to ``fn`` the slice ``t+k``. ``scan`` will follow this logic if partial information is given: * If an output is not wrapped in a dictionary, ``scan`` will wrap it in one assuming that you use only the last step of the output (i.e. it makes your tap value list equal to [-1]). * If you wrap an output in a dictionary and you do not provide any taps but you provide an initial state it will assume that you are using only a tap value of -1. * If you wrap an output in a dictionary but you do not provide any initial state, it assumes that you are not using any form of taps. * If you provide a ``None`` instead of a variable or a empty dictionary ``scan`` assumes that you will not use any taps for this output (like for example in case of a map) If ``outputs_info`` is an empty list or None, ``scan`` assumes that no tap is used for any of the outputs. If information is provided just for a subset of the outputs an exception is raised (because there is no convention on how scan should map the provided information to the outputs of ``fn``) :param non_sequences: ``non_sequences`` is the list of arguments that are passed to ``fn`` at each steps. One can opt to exclude variable used in ``fn`` from this list as long as they are part of the computational graph, though for clarity we encourage not to do so. :param n_steps: ``n_steps`` is the number of steps to iterate given as an int or Theano scalar. If any of the input sequences do not have enough elements, scan will raise an error. If the *value is 0* the outputs will have *0 rows*. If the value is negative, ``scan`` will run backwards in time. If the ``go_backwards`` flag is already set and also ``n_steps`` is negative, ``scan`` will run forward in time. If n stpes is not provided, ``scan`` will figure out the amount of steps it should run given its input sequences. :param truncate_gradient: ``truncate_gradient`` is the number of steps to use in truncated BPTT. If you compute gradients through a scan op, they are computed using backpropagation through time. By providing a different value then -1, you choose to use truncated BPTT instead of classical BPTT, where you go for only ``truncate_gradient`` number of steps back in time. :param go_backwards: ``go_backwards`` is a flag indicating if ``scan`` should go backwards through the sequences. If you think of each sequence as indexed by time, making this flag True would mean that ``scan`` goes back in time, namely that for any sequence it starts from the end and goes towards 0. :param name: When profiling ``scan``, it is crucial to provide a name for any instance of ``scan``. The profiler will produce an overall profile of your code as well as profiles for the computation of one step of each instance of ``scan``. The ``name`` of the instance appears in those profiles and can greatly help to disambiguate information. :param mode: It is recommended to leave this argument to None, especially when profiling ``scan`` (otherwise the results are not going to be accurate). If you prefer the computations of one step of ``scan`` to be done differently then the entire function, you can use this parameter to describe how the computations in this loop are done (see ``theano.function`` for details about possible values and their meaning). :param profile: Flag or string. If true, or different from the empty string, a profile object will be created and attached to the inner graph of scan. In case ``profile`` is True, the profile object will have the name of the scan instance, otherwise it will have the passed string. Profile object collect (and print) information only when running the inner graph with the new cvm linker ( with default modes, other linkers this argument is useless) :rtype: tuple :return: tuple of the form (outputs, updates); ``outputs`` is either a Theano variable or a list of Theano variables representing the outputs of ``scan`` (in the same order as in ``outputs_info``). ``updates`` is a subclass of dictionary specifying the update rules for all shared variables used in scan This dictionary should be passed to ``theano.function`` when you compile your function. The change compared to a normal dictionary is that we validate that keys are SharedVariable and addition of those dictionary are validated to be consistent. """ # Note : see the internal documentation of the scan op for naming # conventions and all other details if options is None: options = {} rvals = scan_utils.canonical_arguments(sequences, outputs_info, non_sequences, go_backwards, n_steps) inputs, states_and_outputs_info, parameters, T = rvals # If we provided a known number of steps ( before compilation) # and if that number is 1 or -1, then we can skip the Scan Op, # and just apply the inner function once # To do that we check here to see the nature of n_steps T_value = None if isinstance(n_steps, (float, int)): T_value = int(n_steps) else: try: T_value = opt.get_scalar_constant_value(n_steps) except (TypeError, AttributeError): T_value = None if T_value in (1, -1): return one_step_scan(fn, inputs, states_and_outputs_info, parameters, truncate_gradient) # 1. Variable representing the current time step t = scalar_shared(numpy.int64(0), name='t') # 2. Allocate memory for the states of scan. mintaps = [] lengths = [] for pos, arg_info in enumerate(states_and_outputs_info): if arg_info.get('taps', None) == [-1]: mintaps.append(1) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) arg_info['initial'] = scan_utils.expand(tensor.unbroadcast( tensor.shape_padleft(arg_info['initial']), 0), T) elif arg_info.get('taps', None): if numpy.any(numpy.array(arg_info.get('taps', [])) > 0): # Make sure we do not have requests for future values of a # sequence we can not provide such values raise ValueError('Can not use future taps of outputs', arg_info) mintap = abs(numpy.min(arg_info['taps'])) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) mintaps.append(mintap) arg_info['initial'] = scan_utils.expand( arg_info['initial'][:mintap], T) else: mintaps.append(0) lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) # 3. Generate arguments for the function passed to scan. This will # function will return the outputs that need to be computed at every # timesteps inputs_slices = [input[t] for input in inputs] states_slices = [] for n, state in enumerate(states_and_outputs_info): # Check if it is actually a state and not an output if mintaps[n] != 0: for k in state['taps']: states_slices.append( state['initial'][(t + mintaps[n] + k) % lengths[n]]) # 4. Construct outputs that are to be computed by the inner # function of scan args = inputs_slices + states_slices + parameters cond, states_and_outputs, updates = \ scan_utils.get_updates_and_outputs(fn(*args)) # User is allowed to provide no information if it only behaves like a # map if (len(states_and_outputs) != len(states_and_outputs_info) and len(states_and_outputs_info) == 0): mintaps = [0] * len(states_and_outputs) # 5. Construct the scan op # 5.1 Construct list of shared variables with updates (those that # can be treated as states (i.e. of TensorType) and those that can not # (like Random States) if cond is not None: _cond = [cond] else: _cond = [] rvals = rebuild_collect_shared( states_and_outputs + _cond, updates=updates, rebuild_strict=True, copy_inputs_over=True, no_default_updates=False) # extracting the arguments input_variables, cloned_outputs, other_rval = rvals clone_d, update_d, update_expr, shared_inputs = other_rval additional_input_states = [] additional_output_states = [] additional_lengths = [] additional_mintaps = [] original_numeric_shared_variables = [] non_numeric_input_states = [] non_numeric_output_states = [] original_non_numeric_shared_variables = [] pos = len(lengths) for sv in shared_inputs: if sv in update_d: if isinstance(sv, (TensorVariable, TensorSharedVariable)): # We can treat it as a sit sot nw_state = scan_utils.expand( tensor.unbroadcast(tensor.shape_padleft(sv), 0), T) additional_lengths.append(scalar_shared(numpy.int64(0), name='l%d' % pos)) pos = pos + 1 additional_mintaps.append(1) additional_input_states.append(nw_state) additional_output_states.append( scan_utils.clone(tensor.set_subtensor( nw_state[(t + 1) % additional_lengths[-1]], update_d[sv]))) original_numeric_shared_variables.append(sv) else: non_numeric_input_states.append(sv) non_numeric_output_states.append(update_d[sv]) original_non_numeric_shared_variables.append(sv) # Replace shared variables in the update _additional_output_states = [] replace = {} for sv, buf in zip(original_numeric_shared_variables, additional_input_states): replace[sv] = buf[t] for out in additional_output_states: _additional_output_states.append( scan_utils.clone(out, replace=replace)) additional_output_states = _additional_output_states # 5.2 Collect inputs/outputs of the inner function inputs = [] outputs = [] for n, mintap in enumerate(mintaps): if mintap != 0: input_state = states_and_outputs_info[n]['initial'] inputs.append(input_state) outputs.append( tensor.set_subtensor( input_state[(t + mintap) % lengths[n]], states_and_outputs[n])) else: mem_buffer = scan_utils.allocate_memory( T, states_and_outputs_info[n], states_and_outputs[n]) inputs.append(output) outputs.append( tensor.set_subtensor(output[t % lengths[n]], states_and_outputs[n])) inputs.extend(additional_input_states) outputs.extend(additional_output_states) lengths.extend(additional_lengths) mintaps.extend(additional_mintaps) inputs.extend(non_numeric_input_states) outputs.extend(non_numeric_output_states) all_other_inputs = gof.graph.inputs(outputs) parameters = [x for x in all_other_inputs if (x not in inputs and x not in lengths and x is not t and isinstance(x, gof.Variable) and not isinstance(x, gof.Constant))] inputs.extend(parameters) # 5.3 Construct the the options dictionary options['name'] = name options['profile'] = profile options['mode'] = mode options['inplace'] = False options['gpu'] = False options['truncate_gradient'] = truncate_gradient options['hash_inner_graph'] = 0 # 5.4 Construct the ScanOp instance local_op = scan_op.ScanOp(inputs=inputs, outputs=outputs, lengths=lengths, switches=[], mintaps=mintaps, index=t, options=options, as_repeatUntil=cond) # Note that we get here all the outputs followed by the update rules to # the shared variables we had in our scan # we know that we have (in this given order): # * len(states_and_outputs) real outputs # * len(additional_input_states) updates for numeric shared variable # * len(non_numeric_input_states) updates for non numeric shared # variables scan_inputs = [T] + inputs scan_outputs_update_rules = scan_utils.to_list(local_op(*scan_inputs)) # 5.5 Collect outputs and add permutation object scan_outputs = [] for pos in xrange(len(states_and_outputs)): out = scan_utils.ScanPermutation(mintaps[pos])( scan_outputs_update_rules[pos], t) scan_outputs.append(out[mintaps[pos]:]) # 5.6 Construct updates dictionary update_rules = scan_outputs_update_rules[len(states_and_outputs):] updates = {} for v, u in izip(original_numeric_shared_variables, update_rules[:len(additional_input_states)]): updates[v] = u[-1] for v, u in izip(original_non_numeric_shared_variables, update_rules[len(additional_input_states):]): updates[v] = u # Step 5.7 We are done and can return everything back to the user return scan_outputs, updates
# No need to print a warning or raise an error now, # it will be done when fn will be called. _logger.info( ('Cannot compute test value for the ' 'inner function of scan, input value missing %s'), e) if getattr(init_out['initial'], 'name', None) is not None: arg.name = init_out['initial'].name + '[t-1]' # We need now to allocate space for storing the output and copy # the initial state over. We do this using the expand function # defined in scan utils sit_sot_scan_inputs.append( scan_utils.expand( tensor.unbroadcast(tensor.shape_padleft(actual_arg), 0), actual_n_steps)) sit_sot_inner_slices.append(actual_arg) if i in return_steps: sit_sot_return_steps[n_sit_sot] = return_steps[i] sit_sot_inner_inputs.append(arg) sit_sot_rightOrder.append(i) n_sit_sot += 1 elif init_out.get('taps', None): if numpy.any(numpy.array(init_out.get('taps', [])) > 0): # Make sure we do not have requests for future values of a # sequence we can not provide such values raise ValueError('Can not use future taps of outputs',
except AttributeError, e: if config.compute_test_value != "ignore": # No need to print a warning or raise an error now, # it will be done when fn will be called. _logger.info( ("Cannot compute test value for the " "inner function of scan, input value missing %s"), e ) if getattr(init_out["initial"], "name", None) is not None: arg.name = init_out["initial"].name + "[t-1]" # We need now to allocate space for storing the output and copy # the initial state over. We do this using the expand function # defined in scan utils sit_sot_scan_inputs.append( scan_utils.expand(tensor.unbroadcast(tensor.shape_padleft(actual_arg), 0), actual_n_steps) ) sit_sot_inner_slices.append(actual_arg) if i in return_steps: sit_sot_return_steps[n_sit_sot] = return_steps[i] sit_sot_inner_inputs.append(arg) sit_sot_rightOrder.append(i) n_sit_sot += 1 elif init_out.get("taps", None): if numpy.any(numpy.array(init_out.get("taps", [])) > 0): # Make sure we do not have requests for future values of a # sequence we can not provide such values raise ValueError("Can not use future taps of outputs", init_out)