Merge pull request #168 from LarrySnyder/sets

Sets

CHANGELOG.md

@@ -9,6 +9,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 - Option to suppress dummy-product indices in simulation output.
+- ``SupplyChainNetwork.nodes_by_index`` dict, which returns a ``SupplyChainNode`` object for a given index.
+Use this in place of ``SupplyChainNetwork.get_node_from_index()``, which will be deprecated.
+
+### Changed
+- Various speedups.
 
 ### Fixed
 - Bug that caused ``mwor_system()`` to crash when ``demand_source`` is provided as argument 

docs/tutorial/tutorial_meio.rst

@@ -49,7 +49,8 @@ You can then add these nodes to a |class_network|:
 
 The network now contains both nodes, and a (directed) edge from ``my_node`` to ``my_other_node``. Nodes can
 be accessed either from a |class_network| object's :attr:`~stockpyl.supply_chain_network.SupplyChainNetwork.nodes` 
-attribute (a list of nodes), or using its member function :meth:`~stockpyl.supply_chain_network.SupplyChainNetwork.get_node_from_index`:
+attribute (a list of nodes), or using its :attr:`~stockpyl.supply_chain_network.SupplyChainNetwork.nodes_by_index`
+attribute (a dict whose keys are indices and whose values are node objects):
 
 .. doctest::
 
@@ -58,7 +59,7 @@ attribute (a list of nodes), or using its member function :meth:`~stockpyl.suppl
 	1
 	>>> n1.local_holding_cost
 	3
-	>>> n2 = network.get_node_from_index(2)
+	>>> n2 = network.nodes_by_index[2]
 	>>> n2.index
 	2
 	>>> n2.echelon_holding_cost

docs/tutorial/tutorial_sim.rst

@@ -516,25 +516,25 @@ methods are the relevant nodes/products, but these arguments can be omitted if t
 
 	.. doctest::
 
-		>>> network.get_node_from_index(1).state_vars[3].get_inventory_level()
+		>>> network.nodes_by_index[1].state_vars[3].get_inventory_level()
 		15
-		>>> network.get_node_from_index(1).state_vars[3].get_order_quantity(predecessor=3)
+		>>> network.nodes_by_index[1].state_vars[3].get_order_quantity(predecessor=3)
 		6
-		>>> network.get_node_from_index(2).state_vars[4].get_inbound_shipment()
+		>>> network.nodes_by_index[2].state_vars[4].get_inbound_shipment()
 		13
-		>>> network.get_node_from_index(2).state_vars[6].get_inbound_order()
+		>>> network.nodes_by_index[2].state_vars[6].get_inbound_order()
 		8
-		>>> network.get_node_from_index(3).state_vars[6].get_inbound_order(successor=2)
+		>>> network.nodes_by_index[3].state_vars[6].get_inbound_order(successor=2)
 		14
 
 Some state variables, in particular those that are not indexed by any node or product,
 are accessed by using the attribute directly:
 
 	.. doctest::
 
-		>>> network.get_node_from_index(1).state_vars[3].holding_cost
+		>>> network.nodes_by_index[1].state_vars[3].holding_cost
 		30
-		>>> network.get_node_from_index(3).state_vars[5].total_cost
+		>>> network.nodes_by_index[3].state_vars[5].total_cost
 		46.0
 
 
@@ -599,7 +599,7 @@ disruptions. First, type-OP disruptions:
 		...     policy_type='BS',
 		...     base_stock_level=[25, 25]
 		... )
-		>>> network.get_node_from_index(2).disruption_process = DisruptionProcess(
+		>>> network.nodes_by_index[2].disruption_process = DisruptionProcess(
 		...     random_process_type='M',
 		...     disruption_type='OP',
 		...     disruption_probability=0.1,
@@ -629,7 +629,7 @@ Next, type-SP disruptions:
 
 	.. doctest::
 
-		>>> network.get_node_from_index(2).disruption_process.disruption_type='SP'
+		>>> network.nodes_by_index[2].disruption_process.disruption_type='SP'
 		>>> _ = simulation(network=network, num_periods=T, rand_seed=42, progress_bar=False)
 		>>> write_results(network=network, num_periods=T, periods_to_print=list(range(7, 16)), columns_to_print=['DISR', 'IO', 'OQ', 'IS', 'OS', 'IL', 'ODI'])
 		  t  | i=1    DISR      IO:2|-2    OQ:EXT|-3    IS:EXT|-3    OS:2|-2    IL:-2    ODI:2|-2  | i=2    DISR      IO:EXT|-4    OQ:1|-2    IS:1|-2    OS:EXT|-4    IL:-4    ODI:EXT|-4
@@ -651,7 +651,7 @@ Next, type-TP disruptions:
 
 	.. doctest::
 
-		>>> network.get_node_from_index(2).disruption_process.disruption_type='TP'
+		>>> network.nodes_by_index[2].disruption_process.disruption_type='TP'
 		>>> _ = simulation(network=network, num_periods=T, rand_seed=42, progress_bar=False)
 		>>> write_results(network=network, num_periods=T, periods_to_print=list(range(7, 16)), columns_to_print=['DISR', 'IO', 'OQ', 'IS', 'ISPL', 'OS', 'IL'])
 		  t  | i=1    DISR      IO:2|-2    OQ:EXT|-3    IS:EXT|-3  ISPL:EXT|-3      OS:2|-2    IL:-2  | i=2    DISR      IO:EXT|-4    OQ:1|-2    IS:1|-2  ISPL:1|-2      OS:EXT|-4    IL:-4
@@ -673,7 +673,7 @@ Finally, type-RP disruptions:
 
 	.. doctest::
 
-		>>> network.get_node_from_index(2).disruption_process.disruption_type='RP'
+		>>> network.nodes_by_index[2].disruption_process.disruption_type='RP'
 		>>> _ = simulation(network=network, num_periods=T, rand_seed=42, progress_bar=False)
 		>>> write_results(network=network, num_periods=T, periods_to_print=list(range(7, 16)), columns_to_print=['DISR', 'IO', 'OQ', 'IS', 'OS', 'IL', 'ISPL', 'IDI'])
 		  t  | i=1    DISR      IO:2|-2    OQ:EXT|-3    IS:EXT|-3  ISPL:EXT|-3      IDI:EXT|-3    OS:2|-2    IL:-2  | i=2    DISR      IO:EXT|-4    OQ:1|-2    IS:1|-2  ISPL:1|-2      IDI:1|-2    OS:EXT|-4    IL:-4

src/stockpyl/aux_files/stockpyl_instances_metadata.rst

@@ -1366,7 +1366,7 @@
 			base_stock_level=[7, 13, 11],
 			initial_inventory_level=[7, 13, 11]
 		)
-		instance.get_node_from_index(0).demand_source.round_to_int = True
+		instance.nodes_by_index[0].demand_source.round_to_int = True
 
 
 .. collapse:: Assembly system from Figure 1 in Rosling (1989)

src/stockpyl/gsm_helpers.py

@@ -191,9 +191,9 @@ def inbound_cst(tree, node_index, cst):
 	for k in node_index:
 		# Determine inbound CST (= max of CST for all predecessors, and external
 		# inbound CST).
-		k_node = tree.get_node_from_index(k)
+		k_node = tree.nodes_by_index[k]
 		SI[k] = k_node.external_inbound_cst
-		if len(k_node.predecessors()) > 0:
+		if len(k_node.predecessor_indices()) > 0:
 			SI[k] = max(SI[k], np.max([cst[i] for i in k_node.predecessor_indices()]))
 
 	if n_is_iterable:
@@ -252,7 +252,7 @@ def net_lead_time(tree, node_index, cst):
 	nlt = {}
 	for k in node_index:
 		# Determine NLT.
-		nlt[k] = SI[k] + tree.get_node_from_index(k).processing_time - cst[k]
+		nlt[k] = SI[k] + tree.nodes_by_index[k].processing_time - cst[k]
 
 	if n_is_iterable:
 		return nlt
@@ -309,7 +309,7 @@ def cst_to_base_stock_levels(tree, node_index, cst):
 
 	base_stock_level = {}
 	for k in node_index:
-		base_stock_level[k] = tree.get_node_from_index(k).net_demand_mean * nlt[k] + ss[k]
+		base_stock_level[k] = tree.nodes_by_index[k].net_demand_mean * nlt[k] + ss[k]
 
 	if n_is_iterable:
 		return base_stock_level
@@ -365,7 +365,7 @@ def safety_stock_levels(tree, node_index, cst):
 
 	safety_stock_level = {}
 	for k in node_index:
-		node_k = tree.get_node_from_index(k)
+		node_k = tree.nodes_by_index[k]
 		safety_stock_level[k] = node_k.demand_bound_constant * \
 						  node_k.net_demand_standard_deviation * \
 						  math.sqrt(nlt[k])

src/stockpyl/gsm_serial.py

@@ -211,11 +211,11 @@ def _cst_dp_serial(network):
 	# Calculate max replenishment times (max replenishment time for node k = SI_N + sum
 	# of processing times at nodes k, ..., N).
 	for k_index in range(num_nodes, 0, -1):
-		k = network.get_node_from_index(k_index)
+		k = network.nodes_by_index[k_index]
 		if k_index == num_nodes:
 			k.max_replenishment_time = k.external_inbound_cst + k.processing_time
 		else:
-			upstream = network.get_node_from_index(k_index + 1)
+			upstream = network.nodes_by_index[k_index + 1]
 			k.max_replenishment_time = upstream.max_replenishment_time + k.processing_time
 
 	# Initialize best_cst_adjacent.
@@ -229,13 +229,13 @@ def _cst_dp_serial(network):
 	best_S = {k_index: {} for k_index in network.node_indices}
 
 	# Get shortcuts to some parameters (for conveience).
-	sigma = network.get_node_from_index(1).demand_source.standard_deviation
+	sigma = network.nodes_by_index[1].demand_source.standard_deviation
 
 	# Loop through stages.
 	for k_index in range(1, num_nodes + 1):
 
 		# Get shortcuts to node (for convenience).
-		k = network.get_node_from_index(k_index)
+		k = network.nodes_by_index[k_index]
 
 		# Determine range of SI values to check. (For node N, it's only external_inbound_cst;
 		# for all other nodes, it's 0, ..., max_replenishment_time - T.)
@@ -290,7 +290,7 @@ def _cst_dp_serial(network):
 	for k_index in range(num_nodes, 0, -1):
 
 		# Get node k.
-		k = network.get_node_from_index(k_index)
+		k = network.nodes_by_index[k_index]
 
 		# Determine SI.
 		if k_index == num_nodes:
@@ -302,7 +302,7 @@ def _cst_dp_serial(network):
 		opt_cst[k_index] = best_S[k_index][SI]
 
 	# Get optimal cost.
-	opt_cost = theta[num_nodes][network.get_node_from_index(num_nodes).external_inbound_cst]
+	opt_cost = theta[num_nodes][network.nodes_by_index[num_nodes].external_inbound_cst]
 
 	return opt_cst, opt_cost
 

src/stockpyl/gsm_tree.py

@@ -175,7 +175,7 @@ def _cst_dp_tree(tree):
 	for k_index in range(min_k_index, max_k_index + 1):
 
 		# Get shortcuts to some parameters (for convenience).
-		k = tree.get_node_from_index(k_index)
+		k = tree.nodes_by_index[k_index]
 		max_replen_time = k.max_replenishment_time
 		proc_time = k.processing_time
 
@@ -225,7 +225,7 @@ def _cst_dp_tree(tree):
 					best_cst_adjacent[k_index][max_replen_time - proc_time]
 
 	# Determine best value of SI for final stage.
-	max_k_node = tree.get_node_from_index(max_k_index)
+	max_k_node = tree.nodes_by_index[max_k_index]
 	SI_dict = {SI: theta_in[max_k_index][SI] for SI in
 			   range(max_k_node.max_replenishment_time -
 					 max_k_node.processing_time + 1)} # smaller range of SI
@@ -245,15 +245,15 @@ def _cst_dp_tree(tree):
 	for k_index in range(max_k_index, min_k_index-1, -1):
 
 		# Get node k.
-		k = tree.get_node_from_index(k_index)
+		k = tree.nodes_by_index[k_index]
 
 		# Get p(k_index), and determine whether p(k_index) and p(p(k_index)) are upstream or
 		# downstream (for convenience).
 		if k_index < max_k_index:
 			pk = k.larger_adjacent_node
 			pk_is_downstream = k.larger_adjacent_node_is_downstream
 			if pk < max_k_index:
-				ppk_is_downstream = tree.get_node_from_index(pk).larger_adjacent_node_is_downstream
+				ppk_is_downstream = tree.nodes_by_index[pk].larger_adjacent_node_is_downstream
 
 		# Where is p(k_index)?
 		if k_index == max_k_index:
@@ -346,7 +346,7 @@ def _calculate_theta_out(tree, k_index, S, theta_in_partial, theta_out_partial):
 	"""
 
 	# Get node k_index, for convenience.
-	k = tree.get_node_from_index(k_index)
+	k = tree.nodes_by_index[k_index]
 
 	# Initialize min_c.
 	min_c = float('inf')
@@ -444,7 +444,7 @@ def _calculate_theta_in(tree, k_index, SI, theta_in_partial, theta_out_partial):
 	"""
 
 	# Get node k_index, for convenience.
-	k = tree.get_node_from_index(k_index)
+	k = tree.nodes_by_index[k_index]
 
 	# Initialize min_c.
 	min_c = float('inf')
@@ -532,7 +532,7 @@ def _calculate_c(tree, k_index, S, SI, theta_in_partial, theta_out_partial):
 	"""
 
 	# Get node k, for convenience.
-	k = tree.get_node_from_index(k_index)
+	k = tree.nodes_by_index[k_index]
 
 	# Initialize output dicts.
 	best_upstream_S = {}
@@ -721,7 +721,7 @@ def relabel_nodes(tree, start_index=0, force_relabel=False):
 	# Fill attributes of relabeled tree.
 	larger_adjacent, downstream = _find_larger_adjacent_nodes(relabeled_tree)
 	for k in tree.nodes:
-		relabeled_node = relabeled_tree.get_node_from_index(new_labels[k.index])
+		relabeled_node = relabeled_tree.nodes_by_index[new_labels[k.index]]
 		relabeled_node.original_label = k.index
 		if new_labels[k.index] < np.max(list(new_labels.values())):
 			relabeled_node.larger_adjacent_node = larger_adjacent[relabeled_node.index]
@@ -900,6 +900,8 @@ def _net_demand(tree):
 	# Make temp copy of tree.
 	temp_tree = copy.deepcopy(tree)
 
+# TODO: add check that the node list gets smaller each iteration -- otherwise there's a bug, but finding it can be hard because of the infinite loop
+
 	# Loop through temp_tree. At each iteration, handle leaf nodes (nodes with
 	# no _successors), adding their net_means and net_variances to those of their
 	# _predecessors. Then remove the leaf nodes and iterate.
@@ -984,7 +986,7 @@ def gsm_to_ssm(tree, p=None):
 		SSM_tree.add_node(SupplyChainNode(n.index, name=n.name, network=SSM_tree,
 			shipment_lead_time=n.processing_time+n.external_inbound_cst,
 			echelon_holding_cost=n.local_holding_cost-upstream_h))
-		SSM_node = SSM_tree.get_node_from_index(n.index)
+		SSM_node = SSM_tree.nodes_by_index[n.index]
 		SSM_node.demand_source = copy.deepcopy(n.demand_source)
 		if p is not None:
 			if n.demand_source is not None:

src/stockpyl/helpers.py

@@ -1319,3 +1319,55 @@ def round_dict_values(the_dict, round_type=None):
 
 	return new_dict
 
+
+### JSON-RELATED FUNCTIONS ###
+
+def serialize_set(obj):
+	"""Serialize a set by converting it to a dict of the form
+	``{'type': 'set', 'elements': elements}``, where ``elements`` are
+	the elements of the set.
+
+	This is used for serializing objects so they can be saved in JSON format. 
+	To use: ``json.dump(json_contents, file, default=serialize_set)``.
+
+	Parameters
+	----------
+	obj : Any
+		The object to serialize.
+
+	Returns
+	-------
+	dict
+		Dictionary representation of ``obj``, if ``obj`` is a set.
+	"""
+	if is_set(obj):
+		return {
+			'type': 'set',
+			'elements': list(obj)
+		}
+
+
+def deserialize_set(obj):
+	"""Deserialize a set of the form ``{'type': 'set', 'elements': elements}``
+	by converting it to a set of the form ``{elements}``.
+		
+	This is used for deserializing objects to they can be loaded from JSON format.
+	To use: ``json.load(file, object_hook=deserialize_set)``.
+
+	Parameters
+	----------
+	obj : Any
+		The object to deserialize.
+
+	Returns
+	-------
+	set
+		Set representation of ``obj``, if ``obj`` is a dict of the appropriate form.
+	"""
+	# https://realpython.com/python-serialize-data/
+	if is_dict(obj) and 'type' in obj and obj['type'] == 'set' and 'elements' in obj:
+		return set(obj['elements'])
+	else:
+		return obj
+
+

src/stockpyl/instances.py

@@ -44,6 +44,8 @@
 
 from stockpyl.supply_chain_network import *
 from stockpyl.supply_chain_node import *
+from stockpyl.helpers import is_set, is_dict, serialize_set, deserialize_set
+
 
 def load_instance(instance_name, filepath=None, ignore_state_vars=True):
 	"""Load an instance from a JSON file. 
@@ -93,7 +95,7 @@ def load_instance(instance_name, filepath=None, ignore_state_vars=True):
 
 	# Load data from JSON.
 	with open(new_path) as f:
-		json_contents = json.load(f)
+		json_contents = json.load(f, object_hook=deserialize_set)
 
 	# Look for instance. (https://stackoverflow.com/a/8653568/3453768)
 	instance_index = next((i for i, item in enumerate(json_contents["instances"]) \
@@ -252,13 +254,12 @@ def save_instance(instance_name, instance_data, instance_description='', filepat
 
 	# Write all instances to JSON.
 	with open(filepath, 'w') as f:
-		json.dump(json_contents, f)
+		json.dump(json_contents, f, default=serialize_set)
 
 	# Close file.
 	f.close()
 
 
-
 def _stockpyl_instances_json_path():
 	"""Determine the path to the JSON file containing the instances. 
 

src/stockpyl/meio_general.py

@@ -280,7 +280,7 @@ def meio_by_coordinate_descent(network, initial_solution=None,
 		if nto_lo[n_ind] is None:
 			nto_lo[n_ind] = 0
 		if nto_hi[n_ind] is None:
-			n = network.get_node_from_index(n_ind)
+			n = network.nodes_by_index[n_ind]
 			nto_hi[n_ind] = 3 * n.lead_time * float(np.sum([s.demand_source.mean for s in network.sink_nodes]))
 
 	# Determine initial solution.