Additional visualization for round 4
This commit is contained in:
vandomej
parent
4efba94ff4
commit
7156d6d733
11 changed files with 721 additions and 129 deletions
4
.gitignore
vendored
4
.gitignore
vendored
|
|
@ -14,3 +14,7 @@ settings.json
|
|||
.DS_Store
|
||||
|
||||
.vscode/alive
|
||||
|
||||
# Added by cargo
|
||||
|
||||
/target
|
||||
|
|
|
|||
|
|
@ -5,10 +5,10 @@ from collections import defaultdict
|
|||
import numpy as np
|
||||
|
||||
# Simplified JSON data for demonstration
|
||||
with open('gemla/round2.json', 'r') as file:
|
||||
with open('gemla/round4.json', 'r') as file:
|
||||
simplified_json_data = json.load(file)
|
||||
|
||||
target_node_id = '0c1e64dc-6ddf-4dbb-bf6e-e8218b925194'
|
||||
target_node_id = '523f8250-3101-4586-90a1-127ffa6d73d9'
|
||||
|
||||
# Function to traverse the tree to find a node id
|
||||
def traverse_left_nodes(node):
|
||||
|
|
@ -67,7 +67,7 @@ fig, ax = plt.subplots(figsize=(10, 6))
|
|||
boxplots = ax.boxplot(scores_values, vert=False, patch_artist=True, labels=[f'Set {i+1}' for i in range(len(scores_values))])
|
||||
|
||||
# Set figure name to node id
|
||||
# fig.canvas.set_window_title('Main node line')
|
||||
ax.set_xscale('symlog', linthresh=1.0)
|
||||
|
||||
# Labeling
|
||||
ax.set_xlabel(f'Scores - Main Line')
|
||||
|
|
@ -79,33 +79,35 @@ ax.set_yticklabels([f'Set {i+1}' for i in range(len(scores_values))])
|
|||
|
||||
# Getting most recent right graph
|
||||
right_nodes = traverse_right_nodes(simplified_json_data[0])
|
||||
target_node_id = None
|
||||
target_node = None
|
||||
if target_node_id:
|
||||
if len(right_nodes) != 0:
|
||||
target_node_id = None
|
||||
target_node = None
|
||||
if target_node_id:
|
||||
for node in right_nodes:
|
||||
if node["val"]["id"] == target_node_id:
|
||||
target_node = node
|
||||
break
|
||||
else:
|
||||
target_node = right_nodes[1]
|
||||
scores = target_node["val"]["node"]["scores"]
|
||||
else:
|
||||
target_node = right_nodes[0]
|
||||
scores = target_node["val"]["node"]["scores"]
|
||||
|
||||
scores_values = [list(score_set.values()) for score_set in scores]
|
||||
scores_values = [list(score_set.values()) for score_set in scores]
|
||||
|
||||
# Set up the figure for plotting on the same graph
|
||||
fig, ax = plt.subplots(figsize=(10, 6))
|
||||
# Set up the figure for plotting on the same graph
|
||||
fig, ax = plt.subplots(figsize=(10, 6))
|
||||
|
||||
# Generate a boxplot for each set of scores on the same graph
|
||||
boxplots = ax.boxplot(scores_values, vert=False, patch_artist=True, labels=[f'Set {i+1}' for i in range(len(scores_values))])
|
||||
# Generate a boxplot for each set of scores on the same graph
|
||||
boxplots = ax.boxplot(scores_values, vert=False, patch_artist=True, labels=[f'Set {i+1}' for i in range(len(scores_values))])
|
||||
|
||||
ax.set_xscale('symlog', linthresh=1.0)
|
||||
|
||||
# Labeling
|
||||
ax.set_xlabel(f'Scores: {target_node['val']['id']}')
|
||||
ax.set_ylabel('Score Sets')
|
||||
ax.yaxis.grid(True) # Add horizontal grid lines for clarity
|
||||
# Labeling
|
||||
ax.set_xlabel(f'Scores: {target_node['val']['id']}')
|
||||
ax.set_ylabel('Score Sets')
|
||||
ax.yaxis.grid(True) # Add horizontal grid lines for clarity
|
||||
|
||||
# Set y-axis labels to be visible
|
||||
ax.set_yticklabels([f'Set {i+1}' for i in range(len(scores_values))])
|
||||
# Set y-axis labels to be visible
|
||||
ax.set_yticklabels([f'Set {i+1}' for i in range(len(scores_values))])
|
||||
|
||||
# Find the highest scoring sets combining all scores and generations
|
||||
scores = []
|
||||
|
|
@ -121,7 +123,8 @@ for node in left_nodes:
|
|||
scores.append(translated_node_scores)
|
||||
|
||||
# Add scores from the right nodes
|
||||
for node in right_nodes:
|
||||
if len(right_nodes) != 0:
|
||||
for node in right_nodes:
|
||||
if node["val"]["node"]:
|
||||
node_scores = node["val"]["node"]["scores"]
|
||||
translated_node_scores = []
|
||||
|
|
@ -155,9 +158,13 @@ labels = [f'{id[:8]}... Gen {gen}' for id, gen in top_20_individual_generations]
|
|||
# Generate box and whisker plots for the top 20 individual generations
|
||||
fig, ax = plt.subplots(figsize=(12, 10))
|
||||
ax.boxplot(top_20_scores, vert=False, patch_artist=True, labels=labels)
|
||||
|
||||
ax.set_xscale('symlog', linthresh=1.0)
|
||||
|
||||
ax.set_xlabel('Scores')
|
||||
ax.set_ylabel('Individual Generation')
|
||||
ax.set_title('Top 20 Individual Generations by Q3 Value')
|
||||
ax.yaxis.grid(True) # Add horizontal grid lines for clarity
|
||||
|
||||
# Display the plot
|
||||
plt.show()
|
||||
|
|
|
|||
9
extract_fann_data/Cargo.toml
Normal file
9
extract_fann_data/Cargo.toml
Normal file
|
|
@ -0,0 +1,9 @@
|
|||
[package]
|
||||
name = "extract_fann_data"
|
||||
version = "0.1.0"
|
||||
edition = "2021"
|
||||
|
||||
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
|
||||
|
||||
[dependencies]
|
||||
fann = "0.1.8"
|
||||
11
extract_fann_data/build.rs
Normal file
11
extract_fann_data/build.rs
Normal file
|
|
@ -0,0 +1,11 @@
|
|||
fn main() {
|
||||
// Replace this with the path to the directory containing `fann.lib`
|
||||
let lib_dir = "F://vandomej/Downloads/vcpkg/packages/fann_x64-windows/lib";
|
||||
|
||||
println!("cargo:rustc-link-search=native={}", lib_dir);
|
||||
println!("cargo:rustc-link-lib=static=fann");
|
||||
// Use `dylib=fann` instead of `static=fann` if you're linking dynamically
|
||||
|
||||
// If there are any additional directories where the compiler can find header files, you can specify them like this:
|
||||
// println!("cargo:include={}", path_to_include_directory);
|
||||
}
|
||||
38
extract_fann_data/src/main.rs
Normal file
38
extract_fann_data/src/main.rs
Normal file
|
|
@ -0,0 +1,38 @@
|
|||
extern crate fann;
|
||||
|
||||
use fann::Fann;
|
||||
use std::os::raw::c_uint;
|
||||
|
||||
fn main() {
|
||||
let args: Vec<String> = std::env::args().collect();
|
||||
if args.len() < 2 {
|
||||
eprintln!("Usage: {} <network_file>", args[0]);
|
||||
std::process::exit(1);
|
||||
}
|
||||
|
||||
let network_file = &args[1];
|
||||
match Fann::from_file(network_file) {
|
||||
Ok(ann) => {
|
||||
// Output layer sizes
|
||||
let layer_sizes = ann.get_layer_sizes();
|
||||
let bias_counts = ann.get_bias_counts();
|
||||
|
||||
println!("Layers:");
|
||||
for (layer_size, bias_count) in layer_sizes.iter().zip(bias_counts.iter()) {
|
||||
println!("{} {}", layer_size, bias_count);
|
||||
}
|
||||
|
||||
// Output connections
|
||||
println!("Connections:");
|
||||
let connections = ann.get_connections();
|
||||
|
||||
for connection in connections {
|
||||
println!("{} {} {}", connection.from_neuron, connection.to_neuron, connection.weight);
|
||||
}
|
||||
},
|
||||
Err(err) => {
|
||||
eprintln!("Error loading network from file {}: {}", network_file, err);
|
||||
std::process::exit(1);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -22,29 +22,39 @@ use std::{
|
|||
ops::Range,
|
||||
path::{Path, PathBuf},
|
||||
};
|
||||
use tokio::{process::Command, sync::mpsc::channel};
|
||||
use tokio::process::Command;
|
||||
use uuid::Uuid;
|
||||
|
||||
use self::neural_network_utility::{crossbreed, major_mutation};
|
||||
|
||||
const BASE_DIR: &str = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations";
|
||||
const POPULATION: usize = 50;
|
||||
const POPULATION: usize = 200;
|
||||
|
||||
const NEURAL_NETWORK_INPUTS: usize = 18;
|
||||
const NEURAL_NETWORK_INPUTS: usize = 22;
|
||||
const NEURAL_NETWORK_OUTPUTS: usize = 8;
|
||||
|
||||
const NEURAL_NETWORK_HIDDEN_LAYERS_MIN: usize = 1;
|
||||
const NEURAL_NETWORK_HIDDEN_LAYERS_MAX: usize = 10;
|
||||
const NEURAL_NETWORK_HIDDEN_LAYERS_MAX: usize = 2;
|
||||
|
||||
const NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MIN: usize = 3;
|
||||
const NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MAX: usize = 35;
|
||||
const NEURAL_NETWORK_INITIAL_WEIGHT_MIN: f32 = -2.0;
|
||||
const NEURAL_NETWORK_INITIAL_WEIGHT_MAX: f32 = 2.0;
|
||||
const NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MAX: usize = 50;
|
||||
|
||||
const NEURAL_NETWORK_INITIAL_WEIGHT_MAX: f32 = 0.5;
|
||||
|
||||
const NEURAL_NETWORK_MINOR_MUTATION_RATE_MAX: f32 = 0.3;
|
||||
const NEURAL_NETWORK_MUTATION_WEIGHT_MAX: f32 = 1.0;
|
||||
|
||||
const NEURAL_NETWORK_MAJOR_MUTATION_RATE_MAX: f32 = 1.0;
|
||||
|
||||
const NEURAL_NETWORK_CROSSBREED_SEGMENTS_MIN: usize = 2;
|
||||
const NEURAL_NETWORK_CROSSBREED_SEGMENTS_MAX: usize = 20;
|
||||
const OFFSHOOT_GENERATIONAL_LENIENCE: u64 = 5;
|
||||
const NEURAL_NETWORK_CROSSBREED_SEGMENTS_MAX: usize = 6;
|
||||
|
||||
const OFFSHOOT_GENERATIONAL_LENIENCE: u64 = 10;
|
||||
const MAINLINE_GENERATIONAL_LENIENCE: u64 = 20;
|
||||
|
||||
const SIMULATION_ROUNDS: usize = 5;
|
||||
const SURVIVAL_RATE: f32 = 0.5;
|
||||
const SURVIVAL_RATE_MIN: f32 = 0.1;
|
||||
const SURVIVAL_RATE_MAX: f32 = 0.9;
|
||||
const GAME_EXECUTABLE_PATH: &str =
|
||||
"F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Package\\Windows\\AI_Fight_Sim.exe";
|
||||
|
||||
|
|
@ -77,6 +87,7 @@ pub struct FighterNN {
|
|||
pub id_mapping: Vec<HashMap<u64, u64>>,
|
||||
pub lerp_amount: f32,
|
||||
pub generational_lenience: u64,
|
||||
pub survival_rate: f32,
|
||||
}
|
||||
|
||||
#[async_trait]
|
||||
|
|
@ -102,9 +113,8 @@ impl GeneticNode for FighterNN {
|
|||
})?;
|
||||
|
||||
let mut nn_shapes = HashMap::new();
|
||||
let weight_initialization_range = thread_rng()
|
||||
.gen_range(NEURAL_NETWORK_INITIAL_WEIGHT_MIN..0.0)
|
||||
..thread_rng().gen_range(0.0..=NEURAL_NETWORK_INITIAL_WEIGHT_MAX);
|
||||
let weight_initialization_amplitude = thread_rng().gen_range(0.0..NEURAL_NETWORK_INITIAL_WEIGHT_MAX);
|
||||
let weight_initialization_range = -weight_initialization_amplitude..weight_initialization_amplitude;
|
||||
|
||||
// Create the first generation in this folder
|
||||
for i in 0..POPULATION {
|
||||
|
|
@ -115,11 +125,11 @@ impl GeneticNode for FighterNN {
|
|||
|
||||
// Randomly generate a neural network shape based on constants
|
||||
let hidden_layers = thread_rng()
|
||||
.gen_range(NEURAL_NETWORK_HIDDEN_LAYERS_MIN..NEURAL_NETWORK_HIDDEN_LAYERS_MAX);
|
||||
.gen_range(NEURAL_NETWORK_HIDDEN_LAYERS_MIN..=NEURAL_NETWORK_HIDDEN_LAYERS_MAX);
|
||||
let mut nn_shape = vec![NEURAL_NETWORK_INPUTS as u32];
|
||||
for _ in 0..hidden_layers {
|
||||
nn_shape.push(thread_rng().gen_range(
|
||||
NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MIN..NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MAX,
|
||||
NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MIN..=NEURAL_NETWORK_HIDDEN_LAYER_SIZE_MAX,
|
||||
) as u32);
|
||||
}
|
||||
nn_shape.push(NEURAL_NETWORK_OUTPUTS as u32);
|
||||
|
|
@ -138,30 +148,31 @@ impl GeneticNode for FighterNN {
|
|||
}
|
||||
|
||||
let mut crossbreed_segments = thread_rng().gen_range(
|
||||
NEURAL_NETWORK_CROSSBREED_SEGMENTS_MIN..NEURAL_NETWORK_CROSSBREED_SEGMENTS_MAX,
|
||||
NEURAL_NETWORK_CROSSBREED_SEGMENTS_MIN..=NEURAL_NETWORK_CROSSBREED_SEGMENTS_MAX,
|
||||
);
|
||||
if crossbreed_segments % 2 == 0 {
|
||||
crossbreed_segments += 1;
|
||||
}
|
||||
|
||||
let mutation_weight_amplitude = thread_rng().gen_range(0.0..1.0);
|
||||
let mutation_weight_amplitude = thread_rng().gen_range(0.0..NEURAL_NETWORK_MUTATION_WEIGHT_MAX);
|
||||
|
||||
Ok(Box::new(FighterNN {
|
||||
id: context.id,
|
||||
folder,
|
||||
population_size: POPULATION,
|
||||
generation: 0,
|
||||
scores: vec![HashMap::new()],
|
||||
scores: vec![],
|
||||
nn_shapes: vec![nn_shapes],
|
||||
// we need crossbreed segments to be even
|
||||
crossbreed_segments,
|
||||
weight_initialization_range,
|
||||
minor_mutation_rate: thread_rng().gen_range(0.0..1.0),
|
||||
major_mutation_rate: thread_rng().gen_range(0.0..1.0),
|
||||
minor_mutation_rate: thread_rng().gen_range(0.0..NEURAL_NETWORK_MINOR_MUTATION_RATE_MAX),
|
||||
major_mutation_rate: thread_rng().gen_range(0.0..NEURAL_NETWORK_MAJOR_MUTATION_RATE_MAX),
|
||||
mutation_weight_range: -mutation_weight_amplitude..mutation_weight_amplitude,
|
||||
id_mapping: vec![HashMap::new()],
|
||||
id_mapping: vec![],
|
||||
lerp_amount: 0.0,
|
||||
generational_lenience: OFFSHOOT_GENERATIONAL_LENIENCE,
|
||||
survival_rate: thread_rng().gen_range(SURVIVAL_RATE_MIN..SURVIVAL_RATE_MAX),
|
||||
}))
|
||||
}
|
||||
|
||||
|
|
@ -189,12 +200,14 @@ impl GeneticNode for FighterNN {
|
|||
i
|
||||
};
|
||||
|
||||
|
||||
let secondary_id = loop {
|
||||
if allotted_simulations.is_empty() {
|
||||
if allotted_simulations.is_empty() || allotted_simulations.len() == 1 {
|
||||
// Select a random id
|
||||
let random_id = loop {
|
||||
let id = thread_rng().gen_range(0..self.population_size);
|
||||
if id != primary_id {
|
||||
allotted_simulations.clear();
|
||||
break id;
|
||||
}
|
||||
};
|
||||
|
|
@ -219,19 +232,18 @@ impl GeneticNode for FighterNN {
|
|||
matches.push((primary_id, secondary_id));
|
||||
}
|
||||
|
||||
debug!("Matches determined");
|
||||
trace!("Matches: {:?}", matches);
|
||||
|
||||
// Create a channel to send the scores back to the main thread
|
||||
let (tx, mut rx) = channel::<(usize, f32)>(self.population_size * SIMULATION_ROUNDS * 20);
|
||||
let mut tasks = Vec::new();
|
||||
|
||||
for (primary_id, secondary_id) in matches.iter() {
|
||||
let task = {
|
||||
let self_clone = self.clone();
|
||||
let semaphore_clone = context.gemla_context.shared_semaphore.clone();
|
||||
let display_simulation_semaphore = context.gemla_context.visible_simulations.clone();
|
||||
let tx = tx.clone();
|
||||
|
||||
let task = async move {
|
||||
let folder = self_clone.folder.clone();
|
||||
let generation = self_clone.generation;
|
||||
|
||||
|
|
@ -245,9 +257,9 @@ impl GeneticNode for FighterNN {
|
|||
.join(self_clone.get_individual_id(*secondary_id as u64))
|
||||
.with_extension("net");
|
||||
|
||||
let permit = semaphore_clone
|
||||
.acquire_owned()
|
||||
.await
|
||||
// Introducing a new scope for acquiring permits and running simulations
|
||||
let simulation_result = async move {
|
||||
let permit = semaphore_clone.acquire_owned().await
|
||||
.with_context(|| "Failed to acquire semaphore permit")?;
|
||||
|
||||
let display_simulation = match display_simulation_semaphore.try_acquire_owned() {
|
||||
|
|
@ -255,66 +267,73 @@ impl GeneticNode for FighterNN {
|
|||
Err(_) => None,
|
||||
};
|
||||
|
||||
let (primary_score, secondary_score) =
|
||||
if let Some(display_simulation) = display_simulation {
|
||||
let (primary_score, secondary_score) = if let Some(display_simulation) = display_simulation {
|
||||
let result = run_1v1_simulation(&primary_nn, &secondary_nn, true).await?;
|
||||
drop(display_simulation);
|
||||
drop(display_simulation); // Explicitly dropping resources no longer needed
|
||||
result
|
||||
} else {
|
||||
run_1v1_simulation(&primary_nn, &secondary_nn, false).await?
|
||||
};
|
||||
|
||||
drop(permit);
|
||||
drop(permit); // Explicitly dropping resources no longer needed
|
||||
|
||||
debug!(
|
||||
"{} vs {} -> {} vs {}",
|
||||
primary_id, secondary_id, primary_score, secondary_score
|
||||
);
|
||||
|
||||
// Send score using a channel
|
||||
tx.send((*primary_id, primary_score))
|
||||
.await
|
||||
.with_context(|| "Failed to send score")?;
|
||||
tx.send((*secondary_id, secondary_score))
|
||||
.await
|
||||
.with_context(|| "Failed to send score")?;
|
||||
Ok((*primary_id, primary_score, *secondary_id, secondary_score))
|
||||
}; // Await the scoped async block immediately
|
||||
|
||||
Ok(())
|
||||
// The result of the simulation, whether Ok or Err, is returned here.
|
||||
// This ensures tx is dropped when the block exits, regardless of success or failure.
|
||||
simulation_result
|
||||
};
|
||||
|
||||
tasks.push(task);
|
||||
}
|
||||
|
||||
let results: Vec<Result<(), Error>> = join_all(tasks).await;
|
||||
debug!("Tasks created");
|
||||
|
||||
let results: Vec<Result<(usize, f32, usize, f32), Error>> = join_all(tasks).await;
|
||||
|
||||
debug!("Tasks completed");
|
||||
|
||||
// resolve results for any errors
|
||||
let mut scores = HashMap::new();
|
||||
for result in results.into_iter() {
|
||||
result.with_context(|| "Failed to run simulation")?;
|
||||
let (primary_id, primary_score, secondary_id, secondary_score) = result.with_context(|| "Failed to run simulation")?;
|
||||
|
||||
// If score exists, add the new score to the existing score
|
||||
if let Some((existing_score, count)) = scores.get_mut(&(primary_id as u64)) {
|
||||
*existing_score += primary_score;
|
||||
*count += 1;
|
||||
} else {
|
||||
scores.insert(primary_id as u64, (primary_score, 1));
|
||||
}
|
||||
|
||||
// Receive the scores from the channel
|
||||
let mut scores = HashMap::new();
|
||||
while let Some((id, score)) = rx.recv().await {
|
||||
// If score exists, add the new score to the existing score
|
||||
if let Some(existing_score) = scores.get_mut(&(id as u64)) {
|
||||
*existing_score += score;
|
||||
if let Some((existing_score, count)) = scores.get_mut(&(secondary_id as u64)) {
|
||||
*existing_score += secondary_score;
|
||||
*count += 1;
|
||||
} else {
|
||||
scores.insert(id as u64, score);
|
||||
scores.insert(secondary_id as u64, (secondary_score, 1));
|
||||
}
|
||||
}
|
||||
|
||||
// Average scores for each individual
|
||||
for (_, score) in scores.iter_mut() {
|
||||
*score /= SIMULATION_ROUNDS as f32;
|
||||
let mut final_scores = HashMap::new();
|
||||
for (i, (score, count)) in scores.iter() {
|
||||
final_scores.insert(*i, *score / *count as f32);
|
||||
}
|
||||
|
||||
self.scores.push(scores);
|
||||
self.scores.push(final_scores);
|
||||
|
||||
Ok(should_continue(&self.scores, self.generational_lenience)?)
|
||||
}
|
||||
|
||||
async fn mutate(&mut self, _context: GeneticNodeContext<Self::Context>) -> Result<(), Error> {
|
||||
let survivor_count = (self.population_size as f32 * SURVIVAL_RATE) as usize;
|
||||
let survivor_count = (self.population_size as f32 * self.survival_rate) as usize;
|
||||
let mut nn_sizes = Vec::new();
|
||||
let mut id_mapping = HashMap::new();
|
||||
|
||||
|
|
@ -359,7 +378,7 @@ impl GeneticNode for FighterNN {
|
|||
let mut tasks = Vec::new();
|
||||
|
||||
// Take the remaining nn's and create new nn's by the following:
|
||||
for i in 0..survivor_count {
|
||||
for i in 0..(self.population_size - survivor_count) {
|
||||
let self_clone = self.clone();
|
||||
|
||||
// randomly select individual id's sorted scores proportional to their score
|
||||
|
|
@ -443,7 +462,6 @@ impl GeneticNode for FighterNN {
|
|||
.collect::<HashMap<_, _>>();
|
||||
|
||||
self.generation += 1;
|
||||
self.scores.push(HashMap::new());
|
||||
self.nn_shapes.push(nn_sizes_map);
|
||||
self.id_mapping.push(id_mapping);
|
||||
|
||||
|
|
@ -524,6 +542,8 @@ impl GeneticNode for FighterNN {
|
|||
run_1v1_simulation(&left_nn_path, &right_nn_path, false).await?
|
||||
};
|
||||
|
||||
debug!("{} vs {} -> {} vs {}", left_nn_id, right_nn_id, left_score, right_score);
|
||||
|
||||
drop(permit);
|
||||
|
||||
Ok::<(f32, f32), Error>((left_score, right_score))
|
||||
|
|
@ -646,21 +666,27 @@ impl GeneticNode for FighterNN {
|
|||
|
||||
debug!("mutation_weight_range: {:?}", mutation_weight_range);
|
||||
|
||||
let survival_rate = left.survival_rate.lerp(right.survival_rate, lerp_amount);
|
||||
|
||||
debug!("survival_rate: {}", survival_rate);
|
||||
|
||||
Ok(Box::new(FighterNN {
|
||||
id: *id,
|
||||
folder,
|
||||
generation: 0,
|
||||
population_size: nn_shapes.len(),
|
||||
scores: vec![HashMap::new()],
|
||||
scores: vec![],
|
||||
crossbreed_segments,
|
||||
nn_shapes: vec![nn_shapes],
|
||||
weight_initialization_range,
|
||||
minor_mutation_rate,
|
||||
major_mutation_rate,
|
||||
mutation_weight_range,
|
||||
id_mapping: vec![HashMap::new()],
|
||||
id_mapping: vec![],
|
||||
lerp_amount,
|
||||
// generational_lenience: left.generational_lenience + MAINLINE_GENERATIONAL_LENIENCE,
|
||||
generational_lenience: MAINLINE_GENERATIONAL_LENIENCE,
|
||||
survival_rate,
|
||||
}))
|
||||
}
|
||||
}
|
||||
|
|
@ -713,7 +739,7 @@ fn should_continue(scores: &[HashMap<u64, f32>], lenience: u64) -> Result<bool,
|
|||
|
||||
debug!(
|
||||
"Highest Q3 value: {} at generation {}, Highest Median value: {} at generation {}, Continuing? {}",
|
||||
highest_q3_value, generation_with_highest_q3, highest_median, generation_with_highest_median, result
|
||||
highest_q3_value, generation_with_highest_q3 + 1, highest_median, generation_with_highest_median + 1, result
|
||||
);
|
||||
|
||||
Ok(result)
|
||||
|
|
@ -767,7 +793,7 @@ async fn run_1v1_simulation(
|
|||
.await
|
||||
.with_context(|| format!("Failed to read score from file: {:?}", score_file))?;
|
||||
|
||||
debug!(
|
||||
trace!(
|
||||
"{} scored {}, while {} scored {}",
|
||||
nn_1_id, round_score, nn_2_id, opposing_score
|
||||
);
|
||||
|
|
@ -790,7 +816,7 @@ async fn run_1v1_simulation(
|
|||
format!("Failed to read score from file: {:?}", opposite_score_file)
|
||||
})?;
|
||||
|
||||
debug!(
|
||||
trace!(
|
||||
"{} scored {}, while {} scored {}",
|
||||
nn_1_id, round_score, nn_2_id, opposing_score
|
||||
);
|
||||
|
|
@ -847,7 +873,7 @@ async fn run_1v1_simulation(
|
|||
.await
|
||||
.with_context(|| format!("Failed to read score from file: {:?}", score_file))?;
|
||||
|
||||
debug!(
|
||||
trace!(
|
||||
"{} scored {}, while {} scored {}",
|
||||
nn_1_id, round_score, nn_2_id, opposing_score
|
||||
);
|
||||
|
|
@ -885,21 +911,17 @@ async fn read_score_from_file(file_path: &Path, nn_id: &str) -> Result<f32, io::
|
|||
"NN ID not found in scores file",
|
||||
));
|
||||
}
|
||||
Err(e)
|
||||
if e.kind() == io::ErrorKind::WouldBlock
|
||||
|| e.kind() == io::ErrorKind::PermissionDenied
|
||||
|| e.kind() == io::ErrorKind::Other =>
|
||||
Err(_) =>
|
||||
{
|
||||
if attempts >= 5 {
|
||||
if attempts >= 2 {
|
||||
// Attempt 5 times before giving up.
|
||||
return Err(e);
|
||||
return Ok(-100.0);
|
||||
}
|
||||
|
||||
attempts += 1;
|
||||
// wait 1 second to ensure the file is written
|
||||
tokio::time::sleep(tokio::time::Duration::from_secs(1)).await;
|
||||
tokio::time::sleep(tokio::time::Duration::from_secs(10)).await;
|
||||
}
|
||||
Err(e) => return Err(e),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -109,9 +109,9 @@ pub fn consolidate_old_connections(
|
|||
|
||||
let primary_shape = primary.get_layer_sizes();
|
||||
let secondary_shape = secondary.get_layer_sizes();
|
||||
debug!("Primary shape: {:?}", primary_shape);
|
||||
debug!("Secondary shape: {:?}", secondary_shape);
|
||||
debug!("New shape: {:?}", new_shape);
|
||||
trace!("Primary shape: {:?}", primary_shape);
|
||||
trace!("Secondary shape: {:?}", secondary_shape);
|
||||
trace!("New shape: {:?}", new_shape);
|
||||
|
||||
// Start by iterating layer by later
|
||||
let primary_connections = primary.get_connections();
|
||||
|
|
|
|||
|
|
@ -6,6 +6,7 @@ use crate::error::Error;
|
|||
|
||||
use anyhow::Context;
|
||||
use async_trait::async_trait;
|
||||
use log::info;
|
||||
use serde::{de::DeserializeOwned, Deserialize, Serialize};
|
||||
use std::fmt::Debug;
|
||||
use uuid::Uuid;
|
||||
|
|
@ -147,6 +148,8 @@ where
|
|||
.await
|
||||
.with_context(|| format!("Error simulating node: {:?}", self))?;
|
||||
|
||||
info!("Simulation complete and continuing: {:?}", next_generation);
|
||||
|
||||
self.state = if next_generation {
|
||||
GeneticState::Mutate
|
||||
} else {
|
||||
|
|
|
|||
380
parameter_analysis.py
Normal file
380
parameter_analysis.py
Normal file
|
|
@ -0,0 +1,380 @@
|
|||
# Re-importing necessary libraries
|
||||
import json
|
||||
import matplotlib.pyplot as plt
|
||||
from collections import defaultdict
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
import seaborn as sns
|
||||
import matplotlib.colors as mcolors
|
||||
import matplotlib.cm as cm
|
||||
import matplotlib.ticker as ticker
|
||||
|
||||
# Simplified JSON data for demonstration
|
||||
with open('gemla/round4.json', 'r') as file:
|
||||
simplified_json_data = json.load(file)
|
||||
|
||||
# Function to traverse the tree to find a node id
|
||||
def traverse_right_nodes(node):
|
||||
if node is None:
|
||||
return []
|
||||
|
||||
right_node = node.get("right")
|
||||
left_node = node.get("left")
|
||||
|
||||
if right_node is None and left_node is None:
|
||||
return []
|
||||
elif right_node and left_node:
|
||||
return [right_node] + traverse_right_nodes(left_node)
|
||||
|
||||
return []
|
||||
|
||||
# Getting most recent right graph
|
||||
right_nodes = traverse_right_nodes(simplified_json_data[0])
|
||||
|
||||
# Heatmaps
|
||||
# Data structure to store mutation rates, generations, and scores
|
||||
mutation_rate_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by mutation rate and generation
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
minor_mutation_rate = node_val["minor_mutation_rate"]
|
||||
generation = node_val["generation"]
|
||||
# Ensure each score is associated with the correct generation
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
mutation_rate_data[minor_mutation_rate][gen_index].append(score)
|
||||
|
||||
# Prepare data for heatmap
|
||||
max_generation = max(max(gens.keys()) for gens in mutation_rate_data.values())
|
||||
heatmap_data = np.full((len(mutation_rate_data), max_generation + 1), np.nan)
|
||||
|
||||
# Populate the heatmap data with average scores
|
||||
mutation_rates = sorted(mutation_rate_data.keys())
|
||||
for i, mutation_rate in enumerate(mutation_rates):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = mutation_rate_data[mutation_rate][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the heatmap
|
||||
df_heatmap = pd.DataFrame(
|
||||
data=heatmap_data,
|
||||
index=mutation_rates,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# Data structure to store major mutation rates, generations, and scores
|
||||
major_mutation_rate_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by major mutation rate and generation
|
||||
# This is assuming the structure to retrieve major_mutation_rate is similar to minor_mutation_rate
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
major_mutation_rate = node_val["major_mutation_rate"]
|
||||
generation = node_val["generation"]
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
major_mutation_rate_data[major_mutation_rate][gen_index].append(score)
|
||||
|
||||
# Prepare the heatmap data for major_mutation_rate similar to minor_mutation_rate
|
||||
major_heatmap_data = np.full((len(major_mutation_rate_data), max_generation + 1), np.nan)
|
||||
major_mutation_rates = sorted(major_mutation_rate_data.keys())
|
||||
|
||||
for i, major_rate in enumerate(major_mutation_rates):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = major_mutation_rate_data[major_rate][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
major_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_major_heatmap = pd.DataFrame(
|
||||
data=major_heatmap_data,
|
||||
index=major_mutation_rates,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# crossbreed_segments
|
||||
# Data structure to store major mutation rates, generations, and scores
|
||||
crossbreed_segments_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by major mutation rate and generation
|
||||
# This is assuming the structure to retrieve major_mutation_rate is similar to minor_mutation_rate
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
crossbreed_segments = node_val["crossbreed_segments"]
|
||||
generation = node_val["generation"]
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
crossbreed_segments_data[crossbreed_segments][gen_index].append(score)
|
||||
|
||||
# Prepare the heatmap data for crossbreed_segments similar to minor_mutation_rate
|
||||
crossbreed_heatmap_data = np.full((len(crossbreed_segments_data), max_generation + 1), np.nan)
|
||||
crossbreed_segments = sorted(crossbreed_segments_data.keys())
|
||||
|
||||
for i, crossbreed_segment in enumerate(crossbreed_segments):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = crossbreed_segments_data[crossbreed_segment][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
crossbreed_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_crossbreed_heatmap = pd.DataFrame(
|
||||
data=crossbreed_heatmap_data,
|
||||
index=crossbreed_segments,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# mutation_weight_range
|
||||
# Data structure to store major mutation rates, generations, and scores
|
||||
mutation_weight_range_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by major mutation rate and generation
|
||||
# This is assuming the structure to retrieve major_mutation_rate is similar to minor_mutation_rate
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
mutation_weight_range = node_val["mutation_weight_range"]
|
||||
positive_extent = mutation_weight_range["end"]
|
||||
negative_extent = -mutation_weight_range["start"]
|
||||
mutation_weight_range = (positive_extent + negative_extent) / 2
|
||||
generation = node_val["generation"]
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
mutation_weight_range_data[mutation_weight_range][gen_index].append(score)
|
||||
|
||||
# Prepare the heatmap data for crossbreed_segments similar to minor_mutation_rate
|
||||
mutation_weight_range_heatmap_data = np.full((len(mutation_weight_range_data), max_generation + 1), np.nan)
|
||||
mutation_weight_ranges = sorted(mutation_weight_range_data.keys())
|
||||
|
||||
for i, mutation_weight_range in enumerate(mutation_weight_ranges):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = mutation_weight_range_data[mutation_weight_range][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
mutation_weight_range_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_mutation_weight_range_heatmap = pd.DataFrame(
|
||||
data=mutation_weight_range_heatmap_data,
|
||||
index=mutation_weight_ranges,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# weight_initialization_range
|
||||
# Data structure to store major mutation rates, generations, and scores
|
||||
weight_initialization_range_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by major mutation rate and generation
|
||||
# This is assuming the structure to retrieve major_mutation_rate is similar to minor_mutation_rate
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
weight_initialization_range = node_val["weight_initialization_range"]
|
||||
positive_extent = weight_initialization_range["end"]
|
||||
negative_extent = -weight_initialization_range["start"]
|
||||
weight_initialization_range = (positive_extent + negative_extent) / 2
|
||||
generation = node_val["generation"]
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
weight_initialization_range_data[weight_initialization_range][gen_index].append(score)
|
||||
|
||||
# Prepare the heatmap data for crossbreed_segments similar to minor_mutation_rate
|
||||
weight_initialization_range_heatmap_data = np.full((len(weight_initialization_range_data), max_generation + 1), np.nan)
|
||||
weight_initialization_ranges = sorted(weight_initialization_range_data.keys())
|
||||
|
||||
for i, weight_initialization_range in enumerate(weight_initialization_ranges):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = weight_initialization_range_data[weight_initialization_range][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
weight_initialization_range_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_weight_initialization_range_heatmap = pd.DataFrame(
|
||||
data=weight_initialization_range_heatmap_data,
|
||||
index=weight_initialization_ranges,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# weight_initialization_range_skew
|
||||
# Data structure to store major mutation rates, generations, and scores
|
||||
weight_initialization_range_skew_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
# Populate the dictionary with scores indexed by major mutation rate and generation
|
||||
# This is assuming the structure to retrieve major_mutation_rate is similar to minor_mutation_rate
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
weight_initialization_range = node_val["weight_initialization_range"]
|
||||
positive_extent = weight_initialization_range["end"]
|
||||
negative_extent = -weight_initialization_range["start"]
|
||||
weight_initialization_range_skew = (positive_extent - negative_extent) / 2
|
||||
generation = node_val["generation"]
|
||||
for gen_index, score_list in enumerate(scores):
|
||||
for score in score_list.values():
|
||||
weight_initialization_range_skew_data[weight_initialization_range_skew][gen_index].append(score)
|
||||
|
||||
# Prepare the heatmap data for crossbreed_segments similar to minor_mutation_rate
|
||||
weight_initialization_range_skew_heatmap_data = np.full((len(weight_initialization_range_skew_data), max_generation + 1), np.nan)
|
||||
weight_initialization_range_skews = sorted(weight_initialization_range_skew_data.keys())
|
||||
|
||||
for i, weight_initialization_range_skew in enumerate(weight_initialization_range_skews):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = weight_initialization_range_skew_data[weight_initialization_range_skew][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
weight_initialization_range_skew_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_weight_initialization_range_skew_heatmap = pd.DataFrame(
|
||||
data=weight_initialization_range_skew_heatmap_data,
|
||||
index=weight_initialization_range_skews,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# Analyze number of neurons correlation to score
|
||||
# We can get the number of neurons via node_val["nn_shapes"] which contains an array of maps
|
||||
# Each map has a key for the individual id and a value which is an array of integers representing the number of neurons in each layer
|
||||
# We can use the individual id to get the score from the scores array
|
||||
# We then generate a density map of the number of neurons vs the score
|
||||
neuron_number_score_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
nn_shapes = node_val["nn_shapes"]
|
||||
# Both scores and nn_shapes are arrays where score is 1 less in length than nn_shapes (each index corresponds to a generation)
|
||||
for gen_index, score in enumerate(scores):
|
||||
for individual_id, nn_shape in nn_shapes[gen_index].items():
|
||||
neuron_number = sum(nn_shape)
|
||||
# check if score has a value for the individual id
|
||||
if individual_id not in score:
|
||||
continue
|
||||
neuron_number_score_data[neuron_number][gen_index].append(score[individual_id])
|
||||
|
||||
# prepare the density map data
|
||||
neuron_number_score_heatmap_data = np.full((len(neuron_number_score_data), max_generation + 1), np.nan)
|
||||
neuron_numbers = sorted(neuron_number_score_data.keys())
|
||||
|
||||
for i, neuron_number in enumerate(neuron_numbers):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = neuron_number_score_data[neuron_number][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
neuron_number_score_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_neuron_number_score_heatmap = pd.DataFrame(
|
||||
data=neuron_number_score_heatmap_data,
|
||||
index=neuron_numbers,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# Analyze number of layers correlation to score
|
||||
nn_layers_score_data = defaultdict(lambda: defaultdict(list))
|
||||
|
||||
for node in right_nodes:
|
||||
node_val = node["val"]["node"]
|
||||
if node_val:
|
||||
scores = node_val["scores"]
|
||||
nn_shapes = node_val["nn_shapes"]
|
||||
# Both scores and nn_shapes are arrays where score is 1 less in length than nn_shapes (each index corresponds to a generation)
|
||||
for gen_index, score in enumerate(scores):
|
||||
for individual_id, nn_shape in nn_shapes[gen_index].items():
|
||||
layer_number = len(nn_shape)
|
||||
# check if score has a value for the individual id
|
||||
if individual_id not in score:
|
||||
continue
|
||||
nn_layers_score_data[layer_number][gen_index].append(score[individual_id])
|
||||
|
||||
# prepare the density map data
|
||||
nn_layers_score_heatmap_data = np.full((len(nn_layers_score_data), max_generation + 1), np.nan)
|
||||
nn_layers = sorted(nn_layers_score_data.keys())
|
||||
|
||||
for i, nn_layer in enumerate(nn_layers):
|
||||
for generation in range(max_generation + 1):
|
||||
scores = nn_layers_score_data[nn_layer][generation]
|
||||
if scores: # Check if there are scores for this generation
|
||||
nn_layers_score_heatmap_data[i, generation] = np.mean(scores)
|
||||
|
||||
# Creating a DataFrame for the major mutation rate heatmap
|
||||
df_nn_layers_score_heatmap = pd.DataFrame(
|
||||
data=nn_layers_score_heatmap_data,
|
||||
index=nn_layers,
|
||||
columns=range(max_generation + 1)
|
||||
)
|
||||
|
||||
# print("Format: ", custom_formatter(0.123498761234, 0))
|
||||
|
||||
# Creating subplots
|
||||
fig, axs = plt.subplots(2, 2, figsize=(20, 14)) # Creates a 3x2 grid of subplots
|
||||
|
||||
# Plotting the minor mutation rate heatmap
|
||||
sns.heatmap(df_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs[0, 0])
|
||||
# axs[0, 0].set_title('Minor Mutation Rate')
|
||||
axs[0, 0].set_xlabel('Minor Mutation Rate')
|
||||
axs[0, 0].set_ylabel('Generation')
|
||||
axs[0, 0].invert_yaxis()
|
||||
|
||||
# Plotting the major mutation rate heatmap
|
||||
sns.heatmap(df_major_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs[0, 1])
|
||||
# axs[0, 1].set_title('Major Mutation Rate')
|
||||
axs[0, 1].set_xlabel('Major Mutation Rate')
|
||||
axs[0, 1].invert_yaxis()
|
||||
|
||||
# Plotting the crossbreed_segments heatmap
|
||||
sns.heatmap(df_crossbreed_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs[1, 0])
|
||||
# axs[1, 0].set_title('Crossbreed Segments')
|
||||
axs[1, 0].set_xlabel('Crossbreed Segments')
|
||||
axs[1, 0].set_ylabel('Generation')
|
||||
axs[1, 0].invert_yaxis()
|
||||
|
||||
# Plotting the mutation_weight_range heatmap
|
||||
sns.heatmap(df_mutation_weight_range_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs[1, 1])
|
||||
# axs[1, 1].set_title('Mutation Weight Range')
|
||||
axs[1, 1].set_xlabel('Mutation Weight Range')
|
||||
axs[1, 1].invert_yaxis()
|
||||
|
||||
fig3, axs3 = plt.subplots(1, 2, figsize=(20, 14)) # Creates a 3x2 grid of subplots
|
||||
|
||||
# Plotting the weight_initialization_range heatmap
|
||||
sns.heatmap(df_weight_initialization_range_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs3[0])
|
||||
# axs[2, 0].set_title('Weight Initialization Range')
|
||||
axs3[0].set_xlabel('Weight Initialization Range')
|
||||
axs3[0].set_ylabel('Generation')
|
||||
axs3[0].invert_yaxis()
|
||||
|
||||
# Plotting the weight_initialization_range_skew heatmap
|
||||
sns.heatmap(df_weight_initialization_range_skew_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs3[1])
|
||||
# axs[2, 1].set_title('Weight Initialization Range Skew')
|
||||
axs3[1].set_xlabel('Weight Initialization Range Skew')
|
||||
axs3[1].set_ylabel('Generation')
|
||||
axs3[1].invert_yaxis()
|
||||
|
||||
# Creating a new window for the scatter plots
|
||||
fig2, axs2 = plt.subplots(2, 1, figsize=(20, 14)) # Creates a 2x1 grid of subplots
|
||||
|
||||
# Plotting the neuron number vs score heatmap
|
||||
sns.heatmap(df_neuron_number_score_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs2[1])
|
||||
# axs[3, 1].set_title('Neuron Number vs. Score')
|
||||
axs2[1].set_xlabel('Neuron Number')
|
||||
axs2[1].set_ylabel('Generation')
|
||||
axs2[1].invert_yaxis()
|
||||
|
||||
# Plotting the number of layers vs score heatmap
|
||||
sns.heatmap(df_nn_layers_score_heatmap.T, cmap='viridis', fmt=".4g", cbar_kws={'label': 'Mean Score'}, ax=axs2[0])
|
||||
# axs[3, 1].set_title('Number of Layers vs. Score')
|
||||
axs2[0].set_xlabel('Number of Layers')
|
||||
axs2[0].set_ylabel('Generation')
|
||||
axs2[0].invert_yaxis()
|
||||
|
||||
# Display the plot
|
||||
plt.tight_layout() # Adjusts the subplots to fit into the figure area.
|
||||
plt.show()
|
||||
118
visualize_networks.py
Normal file
118
visualize_networks.py
Normal file
|
|
@ -0,0 +1,118 @@
|
|||
import matplotlib.pyplot as plt
|
||||
import networkx as nx
|
||||
import subprocess
|
||||
import tkinter as tk
|
||||
from tkinter import filedialog
|
||||
|
||||
def select_file():
|
||||
root = tk.Tk()
|
||||
root.withdraw() # Hide the main window
|
||||
file_path = filedialog.askopenfilename(
|
||||
initialdir="/", # Set the initial directory to search for files
|
||||
title="Select file",
|
||||
filetypes=(("Net files", "*.net"), ("All files", "*.*"))
|
||||
)
|
||||
return file_path
|
||||
|
||||
def get_fann_data(network_file):
|
||||
# Adjust the path to the Rust executable as needed
|
||||
result = subprocess.run(['./extract_fann_data/target/debug/extract_fann_data.exe', network_file], capture_output=True, text=True)
|
||||
if result.returncode != 0:
|
||||
print("Error:", result.stderr)
|
||||
return None, None
|
||||
|
||||
layer_sizes = []
|
||||
connections = []
|
||||
parsing_connections = False
|
||||
|
||||
for line in result.stdout.splitlines():
|
||||
if line.startswith("Layers:"):
|
||||
continue
|
||||
elif line.startswith("Connections:"):
|
||||
parsing_connections = True
|
||||
continue
|
||||
|
||||
if parsing_connections:
|
||||
from_neuron, to_neuron, weight = map(float, line.split())
|
||||
connections.append((int(from_neuron), int(to_neuron), weight))
|
||||
else:
|
||||
layer_size, bias_count = map(int, line.split())
|
||||
layer_sizes.append((layer_size, bias_count))
|
||||
|
||||
return layer_sizes, connections
|
||||
|
||||
def visualize_fann_network(network_file):
|
||||
# Get network data
|
||||
layer_sizes, connections = get_fann_data(network_file)
|
||||
if layer_sizes is None or connections is None:
|
||||
return # Error handling in get_fann_data should provide error output
|
||||
|
||||
# Create a directed graph
|
||||
G = nx.DiGraph()
|
||||
|
||||
# Positions dictionary to hold the position of each neuron
|
||||
pos = {}
|
||||
node_count = 0
|
||||
x_spacing = 1.0
|
||||
y_spacing = 1.0
|
||||
|
||||
# Calculate the maximum layer size for proper spacing
|
||||
max_layer_size = max(size for size, bias in layer_sizes)
|
||||
|
||||
# Build nodes and position them layer by layer from left to right
|
||||
for layer_index, (layer_size, bias_count) in enumerate(layer_sizes):
|
||||
y_positions = list(range(-layer_size-bias_count+1, 1, 1)) # Center-align vertically
|
||||
y_positions = [y * (max_layer_size / (layer_size + bias_count)) * y_spacing for y in y_positions] # Adjust spacing
|
||||
for neuron_index in range(layer_size + bias_count): # Include bias neurons
|
||||
node_label = f"L{layer_index}N{neuron_index}"
|
||||
G.add_node(node_count, label=node_label)
|
||||
pos[node_count] = (layer_index * x_spacing, y_positions[neuron_index % len(y_positions)])
|
||||
node_count += 1
|
||||
|
||||
# Add connections to the graph
|
||||
for from_neuron, to_neuron, weight in connections:
|
||||
G.add_edge(from_neuron, to_neuron, weight=weight)
|
||||
|
||||
max_weight = max(abs(weight) for _, _, weight in connections)
|
||||
print(f"Max weight: {max_weight}")
|
||||
|
||||
# Draw nodes
|
||||
nx.draw_networkx_nodes(G, pos, node_color='skyblue', node_size=200)
|
||||
nx.draw_networkx_labels(G, pos, font_size=7)
|
||||
|
||||
# Custom function for edge properties
|
||||
def adjust_properties(weight):
|
||||
# if weight > 0:
|
||||
# print("Weight:", weight)
|
||||
color = 'green' if weight > 0 else 'red'
|
||||
alpha = min((abs(weight) / max_weight) ** 3, 1)
|
||||
# print(f"Color: {color}, Alpha: {alpha}")
|
||||
return color, alpha
|
||||
|
||||
# Draw edges with custom properties
|
||||
for u, v, d in G.edges(data=True):
|
||||
color, alpha = adjust_properties(d['weight'])
|
||||
nx.draw_networkx_edges(G, pos, edgelist=[(u, v)], edge_color=color, alpha=alpha, width=1.5, arrows=False)
|
||||
|
||||
# Show plot
|
||||
plt.title('FANN Network Visualization')
|
||||
plt.axis('off') # Turn off the axis
|
||||
plt.show()
|
||||
|
||||
# Path to the FANN network file
|
||||
fann_path = 'F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_4f2be613-ab26-4384-9a65-450e043984ea\\6\\4f2be613-ab26-4384-9a65-450e043984ea_fighter_nn_0.net'
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_fc294503-7b2a-40f8-be59-ccc486eb3f79\\0\\fc294503-7b2a-40f8-be59-ccc486eb3f79_fighter_nn_0.net"
|
||||
# fann_path = 'F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_99c30a7f-40ab-4faf-b16a-b44703fdb6cd\\0\\99c30a7f-40ab-4faf-b16a-b44703fdb6cd_fighter_nn_0.net'
|
||||
# Has a 4 layer network
|
||||
# # Generation 1
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98\\1\\16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98_fighter_nn_0.net"
|
||||
# # Generation 5
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98\\5\\16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98_fighter_nn_0.net"
|
||||
# # Generation 10
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98\\10\\16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98_fighter_nn_0.net"
|
||||
# # Generation 20
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98\\20\\16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98_fighter_nn_0.net"
|
||||
# # Generation 32
|
||||
# fann_path = "F:\\\\vandomej\\Projects\\dootcamp-AI-Simulation\\Simulations\\fighter_nn_16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98\\32\\16dfa1b4-03c7-45a6-84b4-22fe3c8e2d98_fighter_nn_0.net"
|
||||
fann_path = select_file()
|
||||
visualize_fann_network(fann_path)
|
||||
|
|
@ -36,7 +36,7 @@ def hierarchy_pos(G, root=None, width=1., vert_gap=0.2, vert_loc=0, xcenter=0.5)
|
|||
return _hierarchy_pos(G, root, width, vert_gap, vert_loc, xcenter)
|
||||
|
||||
# Simplified JSON data for demonstration
|
||||
with open('gemla/round2.json', 'r') as file:
|
||||
with open('gemla/round4.json', 'r') as file:
|
||||
simplified_json_data = json.load(file)
|
||||
|
||||
# Function to traverse the tree and create a graph
|
||||
|
|
@ -68,7 +68,7 @@ def traverse(node, graph, parent=None):
|
|||
# print debug statement
|
||||
# print(f"Node {node_id}: Max score: {overall_max_score:.6f} (Individual {overall_max_score_individual} in Gen {overall_max_score_gen})")
|
||||
# print(f"Left: {node.get('left')}, Right: {node.get('right')}")
|
||||
label = f"{node_id}\nGenerations: {generations}, Population: {population_size}\nMax score: {overall_max_score:.6f} (Individual {overall_max_score_individual} in Gen {overall_max_score_gen})"
|
||||
label = f"{node_id}\nGenerations: {generations}, Population: {population_size}\nMax score: {overall_max_score:.6f} (Individual {overall_max_score_individual} in Gen {overall_max_score_gen + 1 if overall_max_score_gen is not None else 'N/A'})"
|
||||
else:
|
||||
label = node_id
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue