docs: network-hhmodel example: Separate contact from disease progression

[swo]

• Don't use ixa::network, but instead instantiate edges as entities. This allows us to attach the relative rate of transmission to each edge as a property, rather than encoding the edge types as Rust types.
• seir.rs looks for the infected people, then asks network.rs who they contact via get_contacts(). This is a nicer separation of concerns, so that the infection doesn't need to know about the contact structure.
• Make the period of the periodic plan explicit, so that get_contacts() needs a duration argument, that says what contacts will happen over what time period.
• Simplify how the relative rate of transmission gets passed around.
  • Don't specify a "reduction" (that divides); instead specify a relative rate (that gets multiplied).
  • Pass this in certain network::init() to simplify some test interfaces.
  • Attach the relative rate to each edge as a property.

[RobertJacobsonCDC]

Awesome! I think we've all felt the need for someone with modeling expertise to take a close look at our examples for quite some time now, so it's great that you're doing this. If you want to look over any other of the examples—or even add new ones that you think are conspicuously missing—definitely feel free to! But obviously don't feel obligated.

I'll look into the failing lints and get back to you. Sometimes they feel like a moving target...

[RobertJacobsonCDC]

I think it's not failing because of the lints. I think it's failing because the example never terminates: Cancelled after 365m.

[swo]

@bbbruce -- Robert asked for another pair of eyes on this example. Would you take a look? I could do a more full-blown analysis to check for correctness if we consider the prior state to be a gold standard.

[swo]

@bbbruce I reworked a little bit as per our discussion:

• Don't use network module; instead define an Edge entity
• I define Node1 and Node2 as two separate properties. (My first instinct was to define Nodes(node1, node2), but I couldn't figure out how to context.query() for only part of a tuple property, like .query(with!(Edge, Nodes(_, some_person))).)
• I opted for directional edges, which make the queries easier (rather than unidirectional edges, which requires half the entities but makes the queries more complicated).
• I'm having trouble figuring out how to test this. I'd like to say "look for person with data ID 826. They should have 5 contacts." That requires me to (1) query a Person, (2) get their PersonId, and (3) use that PersonId to find the relevant edges. Step 2 has me stumped. Any ideas?

[github-actions]

Benchmark Results

Hyperfine

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
large_sir::baseline	2.9 ± 0.1	2.8	3.1	1.00
large_sir::entities	6.2 ± 0.1	6.0	6.5	2.16 ± 0.06

Criterion

Regressions (slower)

Group	Bench	Param	Change	CI Lower	CI Upper
indexing	query_people_single_indexed_property_entities		26.097%	25.286%	26.717%
indexing	query_people_multiple_individually_indexed_properties_entities		12.586%	12.463%	12.752%
sample_entity	sample_entity_single_property_unindexed	10000	12.131%	10.070%	14.370%
indexing	with_query_results_multiple_individually_indexed_properties_enti		10.619%	10.277%	11.062%
indexing	query_people_indexed_multi-property_entities		9.770%	7.945%	11.650%
indexing	query_people_count_multiple_individually_indexed_properties_enti		7.018%	6.960%	7.077%
sampling	sampling_multiple_known_length_entities		6.159%	5.353%	6.948%
sample_entity	sample_entity_multi_property_indexed	10000	4.509%	4.080%	4.844%
sample_entity	sample_entity_multi_property_indexed	1000	4.074%	3.515%	4.409%
sampling	count_and_sampling_single_known_length_entities		4.058%	3.599%	4.697%
sample_entity	sample_entity_multi_property_indexed	100000	3.652%	2.947%	4.229%
indexing	query_people_count_indexed_multi-property_entities		3.054%	2.302%	3.857%
counts	multi_property_indexed_entities		2.507%	2.295%	2.682%
sampling	sampling_single_known_length_entities		2.493%	2.106%	2.809%
sample_entity	sample_entity_whole_population	10000	2.310%	1.805%	2.875%
large_dataset	bench_query_population_derived_property_entities		1.750%	1.630%	1.866%

Improvements (faster)

Group	Bench	Param	Change	CI Lower	CI Upper
large_dataset	bench_query_population_indexed_property_entities		-24.900%	-26.236%	-23.522%
large_dataset	bench_match_entity		-13.400%	-14.426%	-12.314%
sample_entity	sample_entity_single_property_unindexed	1000	-11.251%	-13.909%	-8.272%
sampling	sampling_single_unindexed_concrete_plus_derived_entities		-6.357%	-6.634%	-6.084%
indexing	query_people_count_single_indexed_property_entities		-5.106%	-5.347%	-4.671%
sampling	count_and_sampling_single_unindexed_concrete_plus_derived_entiti		-4.244%	-4.438%	-4.038%
indexing	with_query_results_single_indexed_property_entities		-3.906%	-4.541%	-3.225%
large_dataset	bench_query_population_multi_indexed_entities		-3.647%	-3.995%	-3.385%
sample_entity	sample_entity_single_property_indexed	1000	-3.174%	-3.728%	-2.747%
sample_entity	sample_entity_single_property_indexed	10000	-2.929%	-3.154%	-2.719%
sampling	sampling_single_unindexed_entities		-1.373%	-1.467%	-1.268%

Unchanged / inconclusive (CI crosses 0%)

Group	Bench	Param	Change	CI Lower	CI Upper
large_dataset	bench_filter_unindexed_entity		5.015%	-0.590%	10.594%
sample_entity	sample_entity_single_property_indexed	100000	-1.818%	-2.685%	-0.828%
sample_entity	sample_entity_whole_population	100000	1.456%	0.888%	2.034%
examples	example-births-deaths		-1.198%	-1.531%	-0.887%
counts	index_after_adding_entities		1.148%	0.921%	1.452%
algorithm_benches	algorithm_sampling_multiple_l_reservoir		1.124%	0.630%	1.556%
sampling	sampling_single_l_reservoir_entities		-1.081%	-1.568%	-0.732%
counts	reindex_after_adding_more_entities		0.709%	0.512%	0.916%
sampling	sampling_multiple_unindexed_entities		-0.687%	-0.902%	-0.385%
large_dataset	bench_query_population_multi_unindexed_entities		-0.619%	-1.574%	0.411%
indexing	with_query_results_indexed_multi-property_entities		-0.589%	-0.956%	-0.253%
sample_entity	sample_entity_single_property_unindexed	100000	0.496%	0.133%	0.862%
counts	concrete_plus_derived_unindexed_entities		0.445%	-0.337%	0.994%
algorithm_benches	algorithm_sampling_single_l_reservoir		-0.402%	-0.687%	-0.148%
examples	example-basic-infection		0.350%	-0.347%	0.835%
sample_entity	sample_entity_whole_population	1000	0.323%	-0.490%	0.877%
sampling	sampling_multiple_l_reservoir_entities		-0.246%	-0.313%	-0.175%
counts	single_property_unindexed_entities		0.181%	-0.623%	1.259%
algorithm_benches	algorithm_sampling_single_known_length		0.163%	-0.446%	1.035%
counts	single_property_indexed_entities		-0.156%	-0.435%	0.103%
algorithm_benches	algorithm_sampling_multiple_known_length		-0.151%	-0.765%	0.260%
counts	multi_property_unindexed_entities		-0.118%	-0.578%	0.374%
algorithm_benches	algorithm_sampling_single_rand_reservoir		0.090%	-0.323%	0.496%
large_dataset	bench_filter_indexed_entity		-0.067%	-12.122%	14.305%
large_dataset	bench_query_population_property_entities		0.038%	-0.661%	0.798%

[swo]

@RobertJacobsonCDC A few changes this round:

• Call it relative_rate and make it clear in the readme what that means
• Clarify some of the language and math around what the relative rate is multiplying
• Simplify function interfaces: mostly pass the relative rate float, rather than the RelativeRate struct
• Use the new system for implementing equality for float properties

[github-actions]

Benchmark Results

Hyperfine

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
large_sir::baseline	3.3 ± 0.0	3.2	3.4	1.01 ± 0.02
large_sir::baseline_households	3.2 ± 0.0	3.2	3.3	1.00
large_sir::entities	7.4 ± 0.1	7.2	7.9	2.29 ± 0.04
large_sir::households	7.1 ± 0.1	6.9	7.4	2.18 ± 0.04

Criterion

Regressions (slower)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Improvements (faster)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Unchanged / inconclusive (CI crosses 0%)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Not Compared (no baseline yet)

Group	Bench	Reason
(none)

[github-actions]

Benchmark Results

Hyperfine

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
large_sir::baseline	2.7 ± 0.1	2.6	3.1	1.01 ± 0.06
large_sir::baseline_households	2.6 ± 0.1	2.6	3.5	1.00
large_sir::entities	5.3 ± 0.2	5.1	6.1	2.01 ± 0.13
large_sir::households	5.0 ± 0.1	4.9	5.3	1.91 ± 0.10

Criterion

Regressions (slower)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Improvements (faster)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Unchanged / inconclusive (CI crosses 0%)

Group	Bench	Param	Change	CI Lower	CI Upper
(none)

Not Compared (no baseline yet)

Group	Bench	Reason
(none)