from plots import EloPlot, Data
import numpy as np
for discount in [0.5, 0.75, 0.95]:
l = [(f'SL-discount-{discount}-NoTrain-alpha-0.000003', 'Simple Linear',
'#9C27B0'),
(f'LA-discount-{discount}-NoTrain-alpha-0.000001',
'Linear Approximation', '#2196F3'),
(f'NN-discount-{discount}-NoTrain-lr-0.0002', 'Neural Network',
'#4CAF50')]
with EloPlot(title=f'Agents, discount={discount} - Elo Rating3',
saves=[f'GoodPlots/Agents-{discount}.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
import pandas as pd
from plots import EloPlot
import numpy as np
data = np.array(
pd.read_csv("outputs/Best-5000/csv/NNAgent4Best-5000-EloOverTime.csv",
header=None))[0, :]
with EloPlot(title=f'The Very Best',
saves=[f'GoodPlots/BestSingle.png']) as plt:
plt.plot(data, label='The Best', color='#4CAF50')
plt.ylim((900, 2520))
plt.doPrint = True
data = np.array(
pd.read_csv("outputs/Fruit-5000/csv/NNAgent15Fruit-5000-EloOverTime.csv",
header=None))[0, :]
with EloPlot(title=f'The Very Best Fruit',
saves=[f'GoodPlots/BestSingleFruit.png'],
CR=False,
CRP=False) as plt:
plt.plot(data, label='The Best Fruit', color='#4CAF50')
plt.doPrint = True
plt.axhline(y=1480.4,
color='#E91E63',
lw=1,
label='CleverRandom+probs',
zorder=1)
plt.axhline(y=1289.0,
color='#9C27B0',
from plots import EloPlot, Data
import numpy as np
l = [('LAMBDA-0.9_DISCOUNT-0.01', 'discount=0.01, lambda=0.9', '#9C27B0'), ('LAMBDA-0.9_DISCOUNT-0.1', 'discount=0.1, lambda=0.9', '#2196F3'), ('LAMBDA-0.9_DISCOUNT-0.5', 'discount=0.5, lambda=0.9', '#4CAF50'), ('LAMBDA-0.9_DISCOUNT-0.9', 'discount=0.9, lambda=0.9', '#F44336'), ('LAMBDA-0.9_dISCOUNT-0.99', 'Discount=0.99, lambda=0.9', '#228B22')]
with EloPlot(title=f'Lambda and discount - lambda - 0.9', saves=[f'GoodPlots/LambdaDiscountLAMBD09.png']) as plt:
for EloOverTime, name, agent, color in Data([f".\\outputs\\{a[0]}\\csv" for a in l], colors=[a[2] for a in l], elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
import numpy as np
l = [('CleverRandomElo-fruit-CalcProb', 'CleverRandom+probs', '#E91E63'),
('CleverRandomElo-fruit', 'CleverRandom', '#9C27B0')]
with EloPlot(title=f'Clever Random - Fruit',
saves=[f'GoodPlots/CleverRandomFruit.png'],
CR=False,
CRP=False) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
theData = EloOverTime[0, 1000:, :]
m = theData.mean()
sd = theData.std() / np.sqrt(theData.size) * 1.96
print(
f'{name}: mean = {m:.6} , 95% = [{(m-sd):.6}, {(m+sd):.6}] Calculated from {theData.shape[1]} samples.'
)
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
with EloPlot(title=f'Probability - Elo Rating',
saves=[f'GoodPlots/Probability.png']) as plt:
for EloOverTime, name, agent, color in Data([
".\\outputs\\4000-calcprob-True\\csv",
".\\outputs\\4000-calcprob-false\\csv"
],
colors=['#4CAF50', '#2196F3'],
elo=False):
name = 'NNAgent+probs' if name == '4000-calcprob-True' else 'NNAgent'
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
import numpy as np
l = [('Best-2000', 'Explorer 2000', '#2196F3'),
('Best-5000', 'Explorer 5000', '#4CAF50')]
with EloPlot(title=f'The Best - Elo Rating',
saves=[f'GoodPlots/BestNoFruit.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
theData = EloOverTime[0, 10000:, :]
m = theData.mean()
sd = theData.std() / np.sqrt(theData.size) * 1.96
print(
f'{name}: mean = {m:.6} , 95% = [{(m-sd):.6}, {(m+sd):.6}] Calculated from {theData.shape[1]} samples.'
)
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, allDirs, Data
maxi, maxiplot = 0, None
for Elo, EloOverTime, name, agent, color in Data(allDirs()):
with EloPlot(xlabel='Players',
title=f'{name} - Elo Rating',
saves=[
f'plots/EloRating/' + f'{name}' + '.png',
f'outputs/{name}/' + f'{name} - Elo Rating' + '.png'
]) as plt:
plt.varPlot(plt, Elo, agent, color)
with EloPlot(
title=f'{name} - Elo Rating over Time',
saves=[
f'plots/EloRatingOverTime/' + f'{name}' + '.png',
f'outputs/{name}/' + f'{name} - Elo Rating over Time' + '.png'
]) as plt:
plt.varPlot(plt, EloOverTime, agent, color)
# plt.maxPlot(plt, EloOverTime, agent, color)
try:
temp = EloOverTime.max()
if temp > maxi:
maxi, maxiplot = temp, name
except:
pass
print(maxi, maxiplot)
for chooser in ['random', 'weighted']:
selfplay = 50
l = [(f'NN-Selfplay-{selfplay}-{chooser}-lr=0.0002-K=2000', 'No Impala',
'#673ab7'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-50-20', 'NN 50 20',
'#9C27B0'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-100-10', 'NN 100 10',
'#2196F3'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-200-5', 'NN 200 5',
'#4CAF50'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-500-2', 'NN 500 20',
'#795548'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-1000-1', 'NN 1000 1',
'#ffeb3b')]
with EloPlot(title=f'Chooserfunction, selfplay={selfplay} - Elo Rating',
saves=[f'GoodPlots/Impala-{chooser}-1000.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
for chooser in ['random', 'weighted']:
selfplay = 50
l = [(f'NN-Selfplay-{selfplay}-{chooser}-lr=0.0002-K=2000-calcprobs',
'No Impala', '#673ab7'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-50-20-calcprobs',
'NN 50 20', '#9C27B0'),
(f'NN-Selfplay-{selfplay}-{chooser}-impala-20-20-100-10-calcprobs',
'NN 100 10', '#2196F3'),
from plots import EloPlot, Data
import numpy as np
l = [('CleverRandomEloCalcProb', 'CleverRandom+probs', '#E91E63'), ('CleverRandomElo', 'CleverRandom', '#9C27B0')]
with EloPlot(title=f'Clever Random - Elo Rating', saves=[f'GoodPlots/CleverRandom.png'], CR=False, CRP=False) as plt:
for EloOverTime, name, agent, color in Data([f".\\outputs\\{a[0]}\\csv" for a in l], colors=[a[2] for a in l], elo=False):
name = [a[1] for a in l if a[0] == name][0]
theData = EloOverTime[0, 1000:, :]
m = theData.mean()
sd = theData.std() / np.sqrt(theData.size) * 1.96
print(f'{name}: mean = {m:.6} , 95% = [{(m-sd):.6}, {(m+sd):.6}] Calculated from {theData.shape[1]} samples.')
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
l = [('4000-Abs', 'Absolut Error', '#2196F3'),
('4000-MME', 'Mean Squared Error', '#4CAF50')]
with EloPlot(title=f'Cost function - Elo Rating',
saves=[f'GoodPlots/Costfunction.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data, colors
import numpy as np
dropouts = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5]
l = [(f'Dropout-{dropout}', f'Dropout {dropout}', colors[i])
for i, dropout in enumerate(dropouts)]
with EloPlot(title=f'Dropout', saves=[f'GoodPlots/Dropout.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
import numpy as np
l = [('4000-Dis-0-lamd-0', 'Discount=0, lambda=0', '#9C27B0'),
('4000-Dis-1-lamd-0', 'Discount=1, lambda=0', '#2196F3'),
('4000-Dis-0-lamd-1', 'Discount=0, lambda=1', '#4CAF50'),
('4000-Dis-1-lamd-1', 'Discount=1, lambda=1', '#F44336')]
with EloPlot(title=f'Lambda and Discount - Elo Rating',
saves=[f'GoodPlots/LambdaDiscount.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
from plots import EloPlot, Data
for selfplay in [20, 50, 100]:
l = [(f'NN-Selfplay-{selfplay}-weighted-lr=0.0002', 'Weighted Chooser',
'#9C27B0'),
(f'NN-Selfplay-{selfplay}-random-lr=0.0002', 'Random Chooser',
'#2196F3'),
(f'NN-Selfplay-{selfplay}-incremental-lr=0.0002',
'Incremental Chooser', '#4CAF50')]
with EloPlot(title=f'Chooserfunction, selfplay={selfplay} - Elo Rating',
saves=[f'GoodPlots/Chooser-{selfplay}.png']) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
plt.varPlot(plt, EloOverTime, name, color)
for selfplay in [20, 50, 100]:
l = [(f'NN-Selfplay-{selfplay}-weighted-lr=0.0002-K=2000',
'Weighted Chooser', '#9C27B0'),
(f'NN-Selfplay-{selfplay}-random-lr=0.0002-K=2000', 'Random Chooser',
'#2196F3'),
(f'NN-Selfplay-{selfplay}-incremental-lr=0.0002-K=2000',
'Incremental Chooser', '#4CAF50')]
with EloPlot(
title=f'Chooserfunction, selfplay={selfplay}, K=2000 - Elo Rating',
from plots import EloPlot, Data
import numpy as np
l = [('Fruit-2000', 'Fruit - Explorer 2000', '#2196F3'),
('Fruit-5000', 'Fruit - Explorer 5000', '#4CAF50')]
with EloPlot(title=f'The Fruit - Elo Rating',
saves=[f'GoodPlots/BestFruit.png'],
CR=False,
CRP=False) as plt:
for EloOverTime, name, agent, color in Data(
[f".\\outputs\\{a[0]}\\csv" for a in l],
colors=[a[2] for a in l],
elo=False):
name = [a[1] for a in l if a[0] == name][0]
theData = EloOverTime[0, 10000:, :]
m = theData.mean()
sd = theData.std() / np.sqrt(theData.size) * 1.96
print(
f'{name}: mean = {m:.6} , 95% = [{(m-sd):.6}, {(m+sd):.6}] Calculated from {theData.shape[1]} samples.'
)
plt.varPlot(plt, EloOverTime, name, color)
plt.axhline(y=1480.4,
color='#E91E63',
lw=1,
label='CleverRandom+probs',
zorder=1)
plt.axhline(y=1289.0,
color='#9C27B0',