Spaces:
Running
Running
A newer version of the Streamlit SDK is available:
1.37.1
!!! note
To run this notebook in JupyterLab, load examples/ex1_0.ipynb
reproduce results from the "InGram" paper
This is an attempt to reproduce the graph of relations example given in lee2023ingram
environment
import os
import pathlib
import typing
from icecream import ic
from pyinstrument import Profiler
import matplotlib.pyplot as plt
import pandas as pd
import pyvis
import textgraphs
%load_ext watermark
%watermark
Last updated: 2024-01-16T17:35:45.550539-08:00
Python implementation: CPython
Python version : 3.10.11
IPython version : 8.20.0
Compiler : Clang 13.0.0 (clang-1300.0.29.30)
OS : Darwin
Release : 21.6.0
Machine : x86_64
Processor : i386
CPU cores : 8
Architecture: 64bit
%watermark --iversions
matplotlib: 3.8.2
pandas : 2.1.4
pyvis : 0.3.2
textgraphs: 0.5.0
sys : 3.10.11 (v3.10.11:7d4cc5aa85, Apr 4 2023, 19:05:19) [Clang 13.0.0 (clang-1300.0.29.30)]
load example graph
load from a JSON file which replicates the data for the "Figure 3" example
graph: textgraphs.GraphOfRelations = textgraphs.GraphOfRelations(
textgraphs.SimpleGraph()
)
ingram_path: pathlib.Path = pathlib.Path(os.getcwd()) / "ingram.json"
graph.load_ingram(
ingram_path,
debug = False,
)
set up the statistical stack profiling
profiler: Profiler = Profiler()
profiler.start()
decouple graph edges into "seeds"
graph.seeds(
debug = True,
)
--- triples in source graph ---
ic| edge.src_node: 0, rel_id: 1, edge.dst_node: 1
ic| edge.src_node: 0, rel_id: 0, edge.dst_node: 2
ic| edge.src_node: 0, rel_id: 0, edge.dst_node: 3
ic| edge.src_node: 4, rel_id: 2, edge.dst_node: 2
ic| edge.src_node: 4, rel_id: 2, edge.dst_node: 3
ic| edge.src_node: 4, rel_id: 1, edge.dst_node: 5
ic| edge.src_node: 6, rel_id: 1, edge.dst_node: 5
ic| edge.src_node: 6, rel_id: 2, edge.dst_node: 7
ic| edge.src_node: 6, rel_id: 4, edge.dst_node: 8
ic| edge.src_node: 9,
Steven_Spielberg Profession Director
Steven_Spielberg Directed Catch_Me_If_Can
Steven_Spielberg Directed Saving_Private_Ryan
Tom_Hanks ActedIn Catch_Me_If_Can
Tom_Hanks ActedIn Saving_Private_Ryan
Tom_Hanks Profession Actor
Mark_Hamil Profession Actor
Mark_Hamil ActedIn Star_Wars
Mark_Hamil BornIn California
rel_id: 5, edge.dst_node: 10
ic| edge.src_node: 9, rel_id: 4, edge.dst_node: 10
ic| edge.src_node: 9, rel_id: 3, edge.dst_node: 8
ic| edge.src_node: 11, rel_id: 4, edge.dst_node: 12
ic| edge.src_node: 11, rel_id: 3, edge.dst_node: 12
ic| edge.src_node: 11, rel_id: 3, edge.dst_node: 8
Brad_Pitt Nationality USA
Brad_Pitt BornIn USA
Brad_Pitt LivedIn California
Clint_Eastwood BornIn San_Francisco
Clint_Eastwood LivedIn San_Francisco
Clint_Eastwood LivedIn California
graph.trace_source_graph()
--- nodes in source graph ---
n: 0, Steven_Spielberg
head: []
tail: [(0, 'Profession', 1), (0, 'Directed', 2), (0, 'Directed', 3)]
n: 1, Director
head: [(0, 'Profession', 1)]
tail: []
n: 2, Catch_Me_If_Can
head: [(0, 'Directed', 2), (4, 'ActedIn', 2)]
tail: []
n: 3, Saving_Private_Ryan
head: [(0, 'Directed', 3), (4, 'ActedIn', 3)]
tail: []
n: 4, Tom_Hanks
head: []
tail: [(4, 'ActedIn', 2), (4, 'ActedIn', 3), (4, 'Profession', 5)]
n: 5, Actor
head: [(4, 'Profession', 5), (6, 'Profession', 5)]
tail: []
n: 6, Mark_Hamil
head: []
tail: [(6, 'Profession', 5), (6, 'ActedIn', 7), (6, 'BornIn', 8)]
n: 7, Star_Wars
head: [(6, 'ActedIn', 7)]
tail: []
n: 8, California
head: [(6, 'BornIn', 8), (9, 'LivedIn', 8), (11, 'LivedIn', 8)]
tail: []
n: 9, Brad_Pitt
head: []
tail: [(9, 'Nationality', 10), (9, 'BornIn', 10), (9, 'LivedIn', 8)]
n: 10, USA
head: [(9, 'Nationality', 10), (9, 'BornIn', 10)]
tail: []
n: 11, Clint_Eastwood
head: []
tail: [(11, 'BornIn', 12), (11, 'LivedIn', 12), (11, 'LivedIn', 8)]
n: 12, San_Francisco
head: [(11, 'BornIn', 12), (11, 'LivedIn', 12)]
tail: []
--- edges in source graph ---
e: 0, Directed
e: 1, Profession
e: 2, ActedIn
e: 3, LivedIn
e: 4, BornIn
e: 5, Nationality
construct a graph of relations
Transform the graph data into graph of relations
graph.construct_gor(
debug = True,
)
ic| node_id: 0, len(seeds
--- transformed triples ---
): 3
ic| trans_arc: TransArc(pair_key=(0, 1),
a_rel=1,
b_rel=0,
node_id=0,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(0, 1),
a_rel=1,
b_rel=0,
node_id=0,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(0, 0),
a_rel=0,
b_rel=0,
node_id=0,
a_dir=<RelDir
(0, 1) Profession.tail Steven_Spielberg Directed.tail
(0, 1) Profession.tail Steven_Spielberg Directed.tail
(0, 0) Directed.tail Steven_Spielberg Directed.tail
.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| node_id: 1, len(seeds
): 1
ic| node_id: 2, len(seeds): 2
ic| trans_arc: TransArc(pair_key=(0, 2),
a_rel=0,
b_rel=2,
node_id=2,
a_dir=<RelDir.HEAD: 0>,
b_dir=<
(0, 2) Directed.head Catch_Me_If_Can ActedIn.head
RelDir.HEAD: 0>)
ic| node_id: 3, len(seeds): 2
ic| trans_arc: TransArc(pair_key=(0, 2),
a_rel=0,
b_rel=2,
node_id=3,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
ic| node_id
(0, 2) Directed.head Saving_Private_Ryan ActedIn.head
: 4, len(seeds): 3
ic| trans_arc: TransArc(pair_key=(2, 2),
a_rel=2,
b_rel=2,
node_id=4,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(1, 2),
a_rel=2,
b_rel=1,
node_id=4,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(1, 2)
(2, 2) ActedIn.tail Tom_Hanks ActedIn.tail
(1, 2) ActedIn.tail Tom_Hanks Profession.tail
(1, 2) ActedIn.tail Tom_Hanks Profession.tail
,
a_rel=2,
b_rel=1,
node_id=4,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic|
node_id: 5, len(seeds): 2
ic| trans_arc: TransArc(pair_key=(1, 1),
a_rel=1,
b_rel=1,
(1, 1) Profession.head Actor Profession.head
node_id=5,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
ic| node_id: 6, len(seeds): 3
ic| trans_arc: TransArc(pair_key=(1, 2),
a_rel=1,
b_rel=2,
node_id=6,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL:
(1, 2) Profession.tail Mark_Hamil ActedIn.tail
1>)
ic| trans_arc: TransArc(pair_key=(1, 4),
a_rel=1,
b_rel=4,
node_id=6,
a_dir
(1, 4) Profession.tail Mark_Hamil BornIn.tail
=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(2, 4),
a_rel=2,
b_rel=4,
node_id=6,
(2, 4) ActedIn.tail Mark_Hamil BornIn.tail
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| node_id: 7, len(seeds): 1
ic| node_id: 8, len(seeds): 3
ic| trans_arc: TransArc(pair_key=(3, 4),
a_rel=4,
b_rel=3,
node_id=8,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD:
(3, 4) BornIn.head California LivedIn.head
0>)
ic| trans_arc: TransArc(pair_key=(3, 4),
a_rel=4,
b_rel=3,
node_id=8,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
ic| trans_arc: TransArc(pair_key=(3, 3),
a_rel=3,
b_rel=3,
node_id=8,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
ic| node_id: 9, len(seeds): 3
ic
(3, 4) BornIn.head California LivedIn.head
(3, 3) LivedIn.head California LivedIn.head
(4, 5) Nationality.tail Brad_Pitt BornIn.tail
| trans_arc: TransArc(pair_key=(4, 5),
a_rel=5,
b_rel=4,
node_id=9,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(3, 5),
a_rel=5,
b_rel=3,
node_id=9,
a_dir=<RelDir.TAIL: 1>,
b_dir=<
(3, 5) Nationality.tail Brad_Pitt LivedIn.tail
RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(3, 4),
a_rel=4,
b_rel=3,
node_id=9,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| node_id: 10, len(seeds): 2
ic| trans_arc: TransArc(pair_key=(4, 5),
a_rel=5,
b_rel=4,
node_id=10,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
ic| node_id: 11, len(seeds): 3
ic| trans_arc: TransArc(pair_key=(3,
(3, 4) BornIn.tail Brad_Pitt LivedIn.tail
(4, 5) Nationality.head USA BornIn.head
(3, 4) BornIn.tail Clint_Eastwood LivedIn.tail
4),
a_rel=4,
b_rel=3,
node_id=11,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic
(3, 4) BornIn.tail Clint_Eastwood LivedIn.tail
| trans_arc: TransArc(pair_key=(3, 4),
a_rel=4,
b_rel=3,
node_id=11,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| trans_arc: TransArc(pair_key=(3, 3),
a_rel=3,
b_rel=3,
node_id=11,
a_dir=<RelDir.TAIL: 1>,
b_dir=<RelDir.TAIL: 1>)
ic| node_id: 12, len(seeds
(3, 3) LivedIn.tail Clint_Eastwood LivedIn.tail
): 2
ic| trans_arc: TransArc(pair_key=(3, 4),
a_rel=4,
b_rel=3,
node_id=12,
a_dir=<RelDir.HEAD: 0>,
b_dir=<RelDir.HEAD: 0>)
(3, 4) BornIn.head San_Francisco LivedIn.head
scores: typing.Dict[ tuple, float ] = graph.get_affinity_scores(
debug = True,
)
--- collect shared entity tallies ---
0 Directed
h: 4 dict_items([(2, 4.0)])
t: 6 dict_items([(0, 3.0), (1, 3.0)])
1 Profession
h: 3 dict_items([(1, 3.0)])
t: 10 dict_items([(0, 3.0), (2, 5.0), (4, 2.0)])
2 ActedIn
h: 4 dict_items([(0, 4.0)])
t: 10 dict_items([(1, 5.0), (2, 3.0), (4, 2.0)])
3 LivedIn
h: 8 dict_items([(3, 3.0), (4, 5.0)])
t: 10 dict_items([(3, 3.0), (4, 5.0), (5, 2.0)])
4 BornIn
h: 7 dict_items([(3, 5.0), (5, 2.0)])
t: 11 dict_items([(1, 2.0), (2, 2.0), (3, 5.0), (5, 2.0)])
5 Nationality
h: 2 dict_items([(4, 2.0)])
t: 4 dict_items([(3, 2.0), (4, 2.0)])
ic(scores);
ic| scores: {(0, 0): 0.3,
(0, 1): 0.2653846153846154,
(0, 2): 0.34285714285714286,
(1, 1): 0.23076923076923078,
(1, 2): 0.3708791208791209,
(1, 4): 0.13247863247863248,
(2, 2): 0.21428571428571427,
(2, 4): 0.12698412698412698,
(3, 3): 0.3333333333333333,
(3, 4): 0.5555555555555556,
(3, 5): 0.2222222222222222,
(4, 5): 0.4444444444444444}
visualize the transform results
graph.render_gor_plt(scores)
plt.show()
pv_graph: pyvis.network.Network = graph.render_gor_pyvis(scores)
pv_graph.force_atlas_2based(
gravity = -38,
central_gravity = 0.01,
spring_length = 231,
spring_strength = 0.7,
damping = 0.8,
overlap = 0,
)
pv_graph.show_buttons(filter_ = [ "physics" ])
pv_graph.toggle_physics(True)
pv_graph.prep_notebook()
pv_graph.show("tmp.fig03.html")
tmp.fig03.html
analysis
As the results below above illustrate, the computed affinity scores differ from what is published in lee2023ingram
. After trying several different variations of interpretation for the paper's descriptions, the current approach provides the closest approximation that we have obtained.
df: pd.DataFrame = graph.trace_metrics(scores)
df
pair | rel_a | rel_b | affinity | expected | |
---|---|---|---|---|---|
0 | (0, 0) | Directed | Directed | 0.30 | NaN |
1 | (0, 1) | Directed | Profession | 0.27 | 0.22 |
2 | (0, 2) | Directed | ActedIn | 0.34 | 0.50 |
3 | (1, 1) | Profession | Profession | 0.23 | NaN |
4 | (1, 2) | Profession | ActedIn | 0.37 | 0.33 |
5 | (1, 4) | Profession | BornIn | 0.13 | 0.11 |
6 | (2, 2) | ActedIn | ActedIn | 0.21 | NaN |
7 | (2, 4) | ActedIn | BornIn | 0.13 | 0.11 |
8 | (3, 3) | LivedIn | LivedIn | 0.33 | NaN |
9 | (3, 4) | LivedIn | BornIn | 0.56 | 0.81 |
10 | (3, 5) | LivedIn | Nationality | 0.22 | 0.11 |
11 | (4, 5) | BornIn | Nationality | 0.44 | 0.36 |
statistical stack profile instrumentation
profiler.stop()
<pyinstrument.session.Session at 0x1416bc7f0>
profiler.print()
_ ._ __/__ _ _ _ _ _/_ Recorded: 17:35:45 Samples: 2526
/_//_/// /_\ / //_// / //_'/ // Duration: 3.799 CPU time: 4.060
/ _/ v4.6.1
Program: /Users/paco/src/textgraphs/venv/lib/python3.10/site-packages/ipykernel_launcher.py -f /Users/paco/Library/Jupyter/runtime/kernel-27f0c564-73f8-45ab-9f64-8b064ae1de10.json
3.799 IPythonKernel.dispatch_queue ipykernel/kernelbase.py:525
└─ 3.791 IPythonKernel.process_one ipykernel/kernelbase.py:511
[10 frames hidden] ipykernel, IPython
3.680 ZMQInteractiveShell.run_ast_nodes IPython/core/interactiveshell.py:3394
├─ 2.176 <module> ../ipykernel_4421/3358887201.py:1
│ └─ 2.176 GraphOfRelations.construct_gor textgraphs/gor.py:311
│ ├─ 1.607 IceCreamDebugger.__call__ icecream/icecream.py:204
│ │ [17 frames hidden] icecream, colorama, ipykernel, thread...
│ │ 1.078 lock.acquire <built-in>
│ └─ 0.566 GraphOfRelations._transformed_triples textgraphs/gor.py:275
│ └─ 0.563 IceCreamDebugger.__call__ icecream/icecream.py:204
│ [13 frames hidden] icecream, colorama, ipykernel, zmq, t...
├─ 0.866 <module> ../ipykernel_4421/4061275008.py:1
│ └─ 0.866 GraphOfRelations.seeds textgraphs/gor.py:197
│ └─ 0.865 IceCreamDebugger.__call__ icecream/icecream.py:204
│ [42 frames hidden] icecream, inspect, posixpath, <built-...
├─ 0.362 <module> ../ipykernel_4421/559531165.py:1
│ ├─ 0.234 show matplotlib/pyplot.py:482
│ │ [32 frames hidden] matplotlib, matplotlib_inline, IPytho...
│ └─ 0.128 GraphOfRelations.render_gor_plt textgraphs/gor.py:522
│ └─ 0.104 draw_networkx networkx/drawing/nx_pylab.py:127
│ [6 frames hidden] networkx, matplotlib
├─ 0.197 <module> ../ipykernel_4421/1169542473.py:1
│ └─ 0.197 IceCreamDebugger.__call__ icecream/icecream.py:204
│ [14 frames hidden] icecream, colorama, ipykernel, thread...
└─ 0.041 <module> ../ipykernel_4421/2247466716.py:1
outro
[ more parts are in progress, getting added to this demo ]