created custom env folder

This commit is contained in:
Ethan Shapiro 2024-03-14 12:39:22 -07:00
parent c121415e31
commit 7ad5b97463
8 changed files with 7188 additions and 1 deletions

3
.gitignore vendored
View File

@ -1 +1,2 @@
**/data/* **/data/*
/env

91
custom_env/agent.py Normal file
View File

@ -0,0 +1,91 @@
import torch
class Agent:
def __init__(self, ) -> None:
# BATCH_SIZE is the number of transitions sampled from the replay buffer
# GAMMA is the discount factor as mentioned in the previous section
# EPS_START is the starting value of epsilon
# EPS_END is the final value of epsilon
# EPS_DECAY controls the rate of exponential decay of epsilon, higher means a slower decay
# TAU is the update rate of the target network
# LR is the learning rate of the ``AdamW`` optimizer
self.batch_size = 128
self.gamma = 0.99
self.eps_start = 0.9
self.eps_end = 0.05
self.eps_decay = 1000
self.tau = 0.005
self.lr = 1e-4
self.n_actions = n_actions
policy_net = DQN(n_observations, n_actions).to(device)
target_net = DQN(n_observations, n_actions).to(device)
target_net.load_state_dict(policy_net.state_dict())
optimizer = optim.AdamW(policy_net.parameters(), lr=LR, amsgrad=True)
memory = ReplayMemory(10000)
def get_state(self, game):
pass
def select_action(state):
sample = random.random()
eps_threshold = EPS_END + (EPS_START - EPS_END) * \
math.exp(-1. * steps_done / EPS_DECAY)
steps_done += 1
if sample > eps_threshold:
with torch.no_grad():
# t.max(1) will return the largest column value of each row.
# second column on max result is index of where max element was
# found, so we pick action with the larger expected reward.
return policy_net(state).max(1).indices.view(1, 1)
else:
return torch.tensor([[env.action_space.sample()]], device=device, dtype=torch.long)
def optimize_model():
if len(memory) < BATCH_SIZE:
return
transitions = memory.sample(BATCH_SIZE)
# Transpose the batch (see https://stackoverflow.com/a/19343/3343043 for
# detailed explanation). This converts batch-array of Transitions
# to Transition of batch-arrays.
batch = Transition(*zip(*transitions))
# Compute a mask of non-final states and concatenate the batch elements
# (a final state would've been the one after which simulation ended)
non_final_mask = torch.tensor(tuple(map(lambda s: s is not None,
batch.next_state)), device=device, dtype=torch.bool)
non_final_next_states = torch.cat([s for s in batch.next_state
if s is not None])
state_batch = torch.cat(batch.state)
action_batch = torch.cat(batch.action)
reward_batch = torch.cat(batch.reward)
# Compute Q(s_t, a) - the model computes Q(s_t), then we select the
# columns of actions taken. These are the actions which would've been taken
# for each batch state according to policy_net
state_action_values = policy_net(state_batch).gather(1, action_batch)
# Compute V(s_{t+1}) for all next states.
# Expected values of actions for non_final_next_states are computed based
# on the "older" target_net; selecting their best reward with max(1).values
# This is merged based on the mask, such that we'll have either the expected
# state value or 0 in case the state was final.
next_state_values = torch.zeros(BATCH_SIZE, device=device)
with torch.no_grad():
next_state_values[non_final_mask] = target_net(non_final_next_states).max(1).values
# Compute the expected Q values
expected_state_action_values = (next_state_values * GAMMA) + reward_batch
# Compute Huber loss
criterion = nn.SmoothL1Loss()
loss = criterion(state_action_values, expected_state_action_values.unsqueeze(1))
# Optimize the model
optimizer.zero_grad()
loss.backward()
# In-place gradient clipping
torch.nn.utils.clip_grad_value_(policy_net.parameters(), 100)
optimizer.step()

View File

@ -0,0 +1,16 @@
import pathlib
import sys
from string import ascii_letters
in_path = pathlib.Path(sys.argv[1])
out_path = pathlib.Path(sys.argv[2])
words = sorted(
{
word.lower()
for word in in_path.read_text(encoding="utf-8").split()
if all(letter in ascii_letters for letter in word)
},
key=lambda word: (len(word), word),
)
out_path.write_text("\n".join(words))

File diff suppressed because it is too large Load Diff

44
custom_env/model.py Normal file
View File

@ -0,0 +1,44 @@
import math
import random
import matplotlib
import matplotlib.pyplot as plt
from collections import namedtuple, deque
from itertools import count
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
Transition = namedtuple('Transition',
('state', 'action', 'next_state', 'reward'))
class ReplayMemory(object):
def __init__(self, capacity: int) -> None:
self.memory = deque([], maxlen=capacity)
def push(self, *args):
self.memory.append(Transition(*args))
def sample(self, batch_size):
return random.sample(self.memory, batch_size)
def __len__(self):
return len(self.memory)
class DQN(nn.Module):
def __init__(self, n_observations: int, n_actions: int) -> None:
super(DQN, self).__init__()
self.layer1 = nn.Linear(n_observations, 128)
self.layer2 = nn.Linear(128, 128)
self.layer3 = nn.Linear(128, n_actions)
def forward(self, x):
x = F.relu(self.layer1(x))
x = F.relu(self.layer2(x))
return self.layer3(x)

61
custom_env/test2.ipynb Normal file
View File

@ -0,0 +1,61 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"from string import ascii_letters, ascii_uppercase, ascii_lowercase"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'ABCDEFGHIJKLMNOPQRSTUVWXYZ'"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"ascii_uppercase"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "env",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.5"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

1098
custom_env/wordlist.txt Normal file

File diff suppressed because it is too large Load Diff

119
custom_env/wyrdl.py Normal file
View File

@ -0,0 +1,119 @@
import contextlib
import pathlib
import random
from string import ascii_letters, ascii_lowercase
from rich.console import Console
from rich.theme import Theme
console = Console(width=40, theme=Theme({"warning": "red on yellow"}))
NUM_LETTERS = 5
NUM_GUESSES = 6
WORDS_PATH = pathlib.Path(__file__).parent / "wordlist.txt"
class Wordle:
def __init__(self) -> None:
self.word_list = WORDS_PATH.read_text(encoding="utf-8").split("\n")
self.n_guesses = 6
self.num_letters = 5
self.curr_word = None
self.reset()
def refresh_page(self, headline):
console.clear()
console.rule(f"[bold blue]:leafy_green: {headline} :leafy_green:[/]\n")
def start_game(self):
# get a new random word
word = self.get_random_word(self.word_list)
self.curr_word = word
def get_state(self):
return
def action_to_word(self, action):
# Calculate the word from the array
word = ''
for i in range(0, len(ascii_lowercase), 26):
# Find the index of 1 in each block of 26
letter_index = action[i:i+26].index(1)
# Append the corresponding letter to the word
word += ascii_lowercase[letter_index]
return word
def play_guess(self, action):
# probably an array of length 26 * 5 for 26 letters and 5 positions
guess = action
def get_random_word(self, word_list):
if words := [
word.upper()
for word in word_list
if len(word) == NUM_LETTERS
and all(letter in ascii_letters for letter in word)
]:
return random.choice(words)
else:
console.print(
f"No words of length {NUM_LETTERS} in the word list",
style="warning",
)
raise SystemExit()
def show_guesses(self, guesses, word):
letter_status = {letter: letter for letter in ascii_lowercase}
for guess in guesses:
styled_guess = []
for letter, correct in zip(guess, word):
if letter == correct:
style = "bold white on green"
elif letter in word:
style = "bold white on yellow"
elif letter in ascii_letters:
style = "white on #666666"
else:
style = "dim"
styled_guess.append(f"[{style}]{letter}[/]")
if letter != "_":
letter_status[letter] = f"[{style}]{letter}[/]"
console.print("".join(styled_guess), justify="center")
console.print("\n" + "".join(letter_status.values()), justify="center")
def guess_word(self, previous_guesses):
guess = console.input("\nGuess word: ").upper()
if guess in previous_guesses:
console.print(f"You've already guessed {guess}.", style="warning")
return guess_word(previous_guesses)
if len(guess) != NUM_LETTERS:
console.print(
f"Your guess must be {NUM_LETTERS} letters.", style="warning"
)
return guess_word(previous_guesses)
if any((invalid := letter) not in ascii_letters for letter in guess):
console.print(
f"Invalid letter: '{invalid}'. Please use English letters.",
style="warning",
)
return guess_word(previous_guesses)
return guess
def reset(self, guesses, word, guessed_correctly, n_episodes):
refresh_page(headline=f"Game: {n_episodes}")
if guessed_correctly:
console.print(f"\n[bold white on green]Correct, the word is {word}[/]")
else:
console.print(f"\n[bold white on red]Sorry, the word was {word}[/]")
if __name__ == "__main__":
main()