Merge pull request #7 from garethjns/policy_agents

Policy agents

README.MD

@@ -96,7 +96,7 @@ python3 -m social-distancing-sim.scripts.run_single_population
 
 ````python
 import social_distancing_sim.environment as env
-# The graph is the "true" population model, containing all the nodes and their data
+# The graph is the "true" environment model, containing all the nodes and their data
 graph = env.Graph(community_n=50,
  community_size_mean=15,
  community_p_in=0.06, # The likelihood of intra-community connections
@@ -159,7 +159,7 @@ def run_and_replay(pop, *args, **kwargs):
 
 save = True
 
-# Create a population with high inter and intra connectivity
+# Create a environment with high inter and intra connectivity
 pop = env.Environment(name='A herd of cats',
  disease=env.Disease(name='COVID-19'),
  observation_space=env.ObservationSpace(graph=env.Graph(community_n=40,
@@ -168,7 +168,7 @@ pop = env.Environment(name='A herd of cats',
  test_rate=1),
  environment_plotting=env.EnvironmentPlotting(ts_fields_g2=["Turn score"]))
 
-# Create a population with reduced inter and intra connectivity
+# Create a environment with reduced inter and intra connectivity
 pop_distanced = env.Environment(name='A socially responsible environment',
  disease=env.Disease(name='COVID-19'),
  observation_space=env.ObservationSpace(graph=env.Graph(community_n=40,
@@ -312,7 +312,6 @@ Parallel(n_jobs=2,
 
 
 # Basic agents and strategy comparison
-![Test basic agents](https://github.com/garethjns/social-distancing-sim/blob/master/images/basic_agents_example.gif)
 
 ````bash
 python3 -m social-distancing-sim.scripts.visual_compare_basic_agents
@@ -379,6 +378,7 @@ Parallel(n_jobs=4,
 ```
 
 # MultiSims: Statistical comparisons - basic agents and strategy comparison
+![Test basic agents](https://github.com/garethjns/social-distancing-sim/blob/master/images/agent_comparison_score_example.png)
 
 ````bash
 python3 -m social-distancing-sim.scripts.stats_compare_basic_agents

scripts/stats_compare_basic_agents.py

@@ -2,8 +2,9 @@
 Run all the basic agents with a number of actions per turn using a MultiSim (n reps = 100). Doesn't save .gifs of
 each rep, rather plots distributions of final scores.
 
-Parameters here match the visual version run in scripts/stats_compare_basic_agents.py.
+Parameters here are similar to the visual version run in scripts/visual_compare_basic_agents.py.
 """
+from typing import List
 
 import matplotlib.pyplot as plt
 import numpy as np
@@ -15,7 +16,8 @@
 import social_distancing_sim.sim as sim
 
 
-def plot_dists(result: str = "Overall score") -> plt.Figure:
+def plot_dists(multi_sims: List[sim.MultiSim],
+ result: str = "Overall score") -> plt.Figure:
  """Plot final score distributions across repetitions, for all agents."""
  fig, axs = plt.subplots(nrows=4,
  ncols=1,
@@ -84,17 +86,17 @@ def plot_dists(result: str = "Overall score") -> plt.Figure:
  n_steps=125)
 
  multi_sims.append(sim.MultiSim(sim_,
- name='basic agent comparison 2',
+ name='basic agent comparison',
  n_reps=100))
 
  # Run all the sims. No need to parallelize here as it's done across n reps in MultiSim.run()
  for ms in tqdm(multi_sims):
  ms.run()
 
- fig = plot_dists("Overall score")
+ fig = plot_dists(multi_sims, "Overall score")
  plt.show()
- fig.savefig('agent_comparison_score.png')
+ fig.savefig('basic_agent_comparison_score.png')
 
- fig = plot_dists("Total deaths")
+ fig = plot_dists(multi_sims, "Total deaths")
  plt.show()
- fig.savefig('agent_comparison_deaths.png')
+ fig.savefig('basic_agent_comparison_deaths.png')

scripts/stats_compare_multi_agents.py

@@ -0,0 +1,131 @@
+"""
+Run all the basic agents with a number of actions per turn using a MultiSim (n reps = 100). Doesn't save .gifs of
+each rep, rather plots distributions of final scores.
+
+Parameters here are similar to the visual version run in scripts/visual_compare_multi_agents.py.
+"""
+from typing import List
+
+import matplotlib.pyplot as plt
+import seaborn as sns
+from tqdm import tqdm
+
+import social_distancing_sim.agent as agent
+import social_distancing_sim.environment as env
+import social_distancing_sim.sim as sim
+
+
+def plot_dists(multi_sims: List[sim.MultiSim],
+ result: str = "Overall score") -> plt.Figure:
+ """Plot final score distributions across repetitions, for all agents."""
+ fig, ax = plt.subplots(nrows=1,
+ ncols=1,
+ figsize=(8, 8))
+
+ min_score = 0
+ max_score = 0
+ for run in multi_sims:
+ min_score = min(min_score, run.results[result].min())
+ max_score = max(max_score, run.results[result].max())
+
+ sns.distplot(run.results[result],
+ hist=False,
+ label=run.sim.agent.name)
+
+ ax.set_title("Policy comparison",
+ fontweight='bold')
+ ax.set_xlim([min_score - abs(min_score * 0.2), max_score + abs(max_score * 0.2)])
+ ax.set_xlabel(ax.get_xlabel(),
+ fontweight='bold')
+ ax.set_ylabel('Prop.',
+ fontweight='bold')
+ ax.legend(title='Agent')
+
+ return fig
+
+
+if __name__ == "__main__":
+ seed = 123
+ steps = 250
+ distancing_params = {"actions_per_turn": 15,
+ "start_step": {'isolate': 15, 'reconnect': 60},
+ "end_step": {'isolate': 55, 'reconnect': steps}}
+ vaccination_params = {"actions_per_turn": 5,
+ "start_step": {'vaccinate': 60},
+ "end_step": {'vaccinate': steps}}
+ treatment_params = {"actions_per_turn": 5,
+ "start_step": {'treat': 50},
+ "end_step": {'treat': steps}}
+
+ # Create a parameter set containing all combinations of the 3 policy agents, and a small set of n_actions
+ agents = [agent.MultiAgent(name="Distancing",
+ agents=[agent.DistancingPolicyAgent(**distancing_params)]),
+ agent.MultiAgent(name="Vaccination",
+ agents=[agent.VaccinationPolicyAgent(**vaccination_params)]),
+ agent.MultiAgent(name="Treatment",
+ agents=[agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Distancing, vaccination",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.VaccinationPolicyAgent(**vaccination_params)]),
+ agent.MultiAgent(name="Distancing, treatment",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Vaccination, treatment",
+ agents=[agent.VaccinationPolicyAgent(**vaccination_params),
+ agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Distancing, vaccination, treatment",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.VaccinationPolicyAgent(**vaccination_params),
+ agent.TreatmentPolicyAgent(**treatment_params)])]
+
+ # Loop over the parameter set and create the Agents, Environments, and the Sim handler
+ multi_sims = []
+ for agt in agents:
+ # Name the environment according to the agent used
+ env_ = env.Environment(name=f"{type(agt).__name__} - {agt.name}",
+ action_space=env.ActionSpace(vaccinate_cost=0,
+ treat_cost=0,
+ isolate_cost=0,
+ isolate_efficiency=0.70,
+ reconnect_efficiency=0.2,
+ treatment_conclusion_chance=0.5,
+ treatment_recovery_rate_modifier=1.8,
+ vaccinate_efficiency=1.25),
+ disease=env.Disease(name='COVID-19',
+ virulence=0.0055,
+ seed=None,
+ immunity_mean=0.7,
+ recovery_rate=0.9,
+ immunity_decay_mean=0.01),
+ healthcare=env.Healthcare(capacity=200),
+ observation_space=env.ObservationSpace(
+ graph=env.Graph(community_n=30,
+ community_size_mean=20,
+ community_p_out=0.08,
+ community_p_in=0.16,
+ seed=None),
+ test_rate=1,
+ seed=None),
+ initial_infections=5,
+ random_infection_chance=1,
+ seed=None)
+
+ sim_ = sim.Sim(env=env_,
+ agent=agt,
+ n_steps=150)
+
+ multi_sims.append(sim.MultiSim(sim_,
+ name='policy agent comparison',
+ n_reps=100))
+
+ # Run all the sims. No need to parallelize here as it's done across n reps in MultiSim.run()
+ for ms in tqdm(multi_sims):
+ ms.run()
+
+ fig = plot_dists(multi_sims, "Overall score")
+ plt.show()
+ fig.savefig('multi_agent_comparison_score.png')
+
+ fig = plot_dists(multi_sims, "Total deaths")
+ plt.show()
+ fig.savefig('multi_agent_comparison_deaths.png')

scripts/visual_compare_basic_agents_small.py

@@ -1,9 +1,4 @@
-
-"""
-Run all the basic agents with a number of actions per turn (n reps = 1). Generate and save .gif.
-
-Parameters here match the stats version run in scripts/stats_compare_basic_agents.py.
-"""
+"""Run all the basic agents with a number of actions per turn (n reps = 1). Generate and save .gif."""
 
 import numpy as np
 from joblib import Parallel, delayed
@@ -62,4 +57,3 @@ def run_and_replay(sim):
  # Run all the prepared Sims
  Parallel(n_jobs=4,
  backend='loky')(delayed(run_and_replay)(sim) for sim in sims)
-

scripts/visual_compare_multi_agents.py

@@ -0,0 +1,92 @@
+"""A number of different MultiAgent setups (n reps = 1). Generate and save .gif."""
+
+from joblib import Parallel, delayed
+
+import social_distancing_sim.agent as agent
+import social_distancing_sim.environment as env
+import social_distancing_sim.sim as sim
+
+
+def run_and_replay(sim):
+ sim.run()
+ if sim.save:
+ sim.env.replay()
+
+
+if __name__ == "__main__":
+ seed = 123
+ steps = 250
+ distancing_params = {"actions_per_turn": 15,
+ "start_step": {'isolate': 15, 'reconnect': 60},
+ "end_step": {'isolate': 55, 'reconnect': steps}}
+ vaccination_params = {"actions_per_turn": 5,
+ "start_step": {'vaccinate': 60},
+ "end_step": {'vaccinate': steps}}
+ treatment_params = {"actions_per_turn": 5,
+ "start_step": {'treat': 50},
+ "end_step": {'treat': steps}}
+
+ # Create a parameter set containing all combinations of the 3 policy agents, and a small set of n_actions
+ agents = [agent.MultiAgent(name="Distancing",
+ agents=[agent.DistancingPolicyAgent(**distancing_params)]),
+ agent.MultiAgent(name="Vaccination",
+ agents=[agent.VaccinationPolicyAgent(**vaccination_params)]),
+ agent.MultiAgent(name="Treatment",
+ agents=[agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Distancing, vaccination",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.VaccinationPolicyAgent(**vaccination_params)]),
+ agent.MultiAgent(name="Distancing, treatment",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Vaccination, treatment",
+ agents=[agent.VaccinationPolicyAgent(**vaccination_params),
+ agent.TreatmentPolicyAgent(**treatment_params)]),
+ agent.MultiAgent(name="Distancing, vaccination, treatment",
+ agents=[agent.DistancingPolicyAgent(**distancing_params),
+ agent.VaccinationPolicyAgent(**vaccination_params),
+ agent.TreatmentPolicyAgent(**treatment_params)])]
+
+ # Loop over the parameter set and create the Agents, Environments, and the Sim handler
+ sims = []
+ for agt in agents:
+ # Name the environment according to the agent used
+ env_ = env.Environment(name=f"{type(agt).__name__} - {agt.name}",
+ action_space=env.ActionSpace(isolate_efficiency=0.75,
+ reconnect_efficiency=0.2,
+ treatment_conclusion_chance=0.2,
+ treatment_recovery_rate_modifier=1.8,
+ vaccinate_efficiency=0.95),
+ disease=env.Disease(name='COVID-19',
+ virulence=0.0055,
+ seed=seed,
+ immunity_mean=0.7,
+ recovery_rate=0.95,
+ immunity_decay_mean=0.012),
+ healthcare=env.Healthcare(capacity=200),
+ environment_plotting=env.EnvironmentPlotting(
+ auto_lim_x=False,
+ ts_fields_g2=["Actions taken", "Vaccinate actions", "Isolate actions",
+ "Reconnect actions", "Treat actions"]),
+ observation_space=env.ObservationSpace(
+ graph=env.Graph(community_n=30,
+ community_size_mean=20,
+ community_p_out=0.08,
+ community_p_in=0.16,
+ seed=seed + 1),
+ test_rate=1,
+ seed=seed + 2),
+ initial_infections=5,
+ random_infection_chance=1,
+ seed=seed + 3)
+
+ sims.append(sim.Sim(env=env_,
+ agent=agt,
+ n_steps=steps,
+ plot=False,
+ save=True,
+ tqdm_on=True)) # Show progress bars for running sims
+
+ # Run all the prepared Sims
+ Parallel(n_jobs=9,
+ backend='loky')(delayed(run_and_replay)(sim) for sim in sims)