xiaohan2012 / active-cascade-reconstruction

sampling-steiner-tree

get started

We made a Docker image that provides the running environment, you can pull from the Docker:

> docker pull xiaohan2012/active_cascade_reconstruction

Then, you can enter the container's shell environment:

> cd active_cascade_reconstruction
> ./start_docker.sh
# (now, you are in docker container)

To test the code,

> cd /code/active_cascade_reconstruction
> pytest test*.py

algorithm files

• query selection algorithms
  • query_selection.py: random, pagerank, entropy, prediction_error
  • tree_stat.py: query score calculation for entropy and prediction_error,
  • sample_pool.py: steiner tree sampling part (including incrementally sample update upon queries)
• estimating infection probability
  • inference.py: also calls functions in tree_stat.py

main experiment scripts

• make_graph_weighted.py: dump graph with edge weights
• simulate_cascades.py: simulate cascades and dump to files
• generate_queries.py: [batch] produce queries and dump to files
  • query_one_round.py: [single] produce queries and dump to file
• infer_from_queries.py: [in batch] infer the infection probabilities from queries
  • infer_one_round.py: [single]
• query_process_illustration.py: visualize different query strategies

typical pipeline to run experiment

1. ./scripts/gen_cascade.sh: generate the cascades
2. ./scripts/run.sh: generate queries and infer the infection proability
3. ./scripts/eval_plot.sh: performance evaluation and plotting

pipeline on Triton

1. edit the configuration file under exp_configs/
2. python3 runner.py --name {name_of_the_config_file_editted_before}
   • the sbatch file path will be printed as ${path_of_sbatch_file}
3. sbatch ${path_of_sbatch_file} to submit the job
   • the result will be saved to postgres DB
4. optionally, you can view the result in the database

plotting scripts

• sample_size_effect_plot.py: inference performance by effect of sample size using different sampling methods
• compare_query_methods_plot.py: comparing different query strategies

note on Triton

use ./singularity/exec.sh ${cmd_to_exec} to execute the command in the singularity container

---

possibly deprecated

shell scripts

graph preprocessing

• use make_graph_weighted.py to add weights and use the resulting graph to simulate cascade
• use preprocess_graph.py to apply global normalization and edge reversing

cascade generation

• gen_cascades_with_varying_size.sh
• gen_cascades_with_varying_obs_fraction.sh

UPDATE Mar 8

for IC, cascade size depends on the edge probability. so setting cascade size needs to be done by trying different edge probabilities

to achieve so, do the following

1. generate the weighted graph using make_weighted_graph.py
2. generate cascade using ./scripts/gen_cascade_weighted.sh
3. check the cascade size using print_cascade_sizes.py
4. modify the edge weights and go back to 1
5. once you are fine with the size, edit and run ./scripts/mv_graph_and_cascade.sh

query

• run.sh: run the query selection as well as inference part
• query_cascades_with_varying_size.sh
• query_cascades_with_varying_obs_fraction.sh
• query_weighted_graph.sh: on weighted graph

note on the directories:

• outputs/queries-weighted: for the weighted graph
  • it contain queries using unweighted sampling, then the subdirectory has name {entropy, prediction_error}-unweighted
• outputs/queries: for the unweighted graph

inference

• infer_weighted_cascade.sh: on weighted graph
• infer_cascades_with_varying_size.sh: TBD
• infer_cascades_with_varying_obs_fraction.sh: TBD

note on the directories:

• outputs/inf_probas-weighted: inference using weighted sampling
• outputs/inf_probas-unweighted: inference using unweighted sampling

misc

• scripts/query_{dataset}.sh: generate queries for ${dataset}
• scripts/infer_{dataset}.sh: make inference on infections for ${dataset}
• compare_running_time.py: script to compare running time for different query methods
• print_cascade_sizes.py: print simulated cascade statistics