-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathendgame_nn_colab.py
509 lines (420 loc) · 91 KB
/
endgame_nn_colab.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
# -*- coding: utf-8 -*-
"""endgame_nn_colab.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1fi4JHdzsMMnlZph-d5wg-3JsB69e6hmM
Note: you need to upload the training data and/or syzygy files to the notebook before running this code
"""
!pip install chess
#import chess
#print(chess.__version__)
# Import relevant modules
from matplotlib import pyplot as plt
import numpy as np
import chess
import chess.syzygy
import random
import tensorflow as tf
import pandas as pd
#import import_data as imp
#import train_model as train
#import define_model as defmod
#import graph as gr
#import gen_pos as gp
print(chess.__version__)
"""Graph.py"""
# this file is where graphing functions will go
# We can use this to plot the training loss over time
def plot_curve(epochs, hist, list_of_metrics):
"""Plot a curve of one or more classification metrics vs. epoch."""
# list_of_metrics should be one of the names shown in:
# https://www.tensorflow.org/tutorials/structured_data/imbalanced_data#define_the_model_and_metrics
plt.figure()
plt.xlabel("Epoch")
plt.ylabel("Value")
for m in list_of_metrics:
x = hist[m]
plt.plot(epochs[1:], x[1:], label=m)
plt.legend()
plt.show()
"""gen_pos.py"""
def gen_fen(material):
"""Return a board with the specific material balance"""""
# Step 1: Create a blank board with either white or black to move
# We are assuming no castling or ep capture will be possible
if random.randint(0, 1) == 1:
board = chess.Board('8/8/8/8/8/8/8/8 w - - 0 1')
else:
board = chess.Board('8/8/8/8/8/8/8/8 b - - 0 1')
# Step 2: Decide whether the first group of material is for white or black
if random.randint(0, 1) == 1:
material = material.swapcase()
# Step 3: Loop over material and add it to the board
dest_squares = random.sample(range(64), len(material))
for piece in material:
location = dest_squares.pop()
board.set_piece_at(location, chess.Piece.from_symbol(piece))
# Step 4: Check that the position is valid
if not board.is_valid() or board.is_checkmate() or board.is_stalemate():
# Recursively get a new try
board = gen_fen(material)
return board
def is_white_square(square):
"""Returns True if the square is on the 'white' diagonals"""
file = chess.square_file(square)
rank = chess.square_rank(square)
if (file + rank) % 2 == 0:
return 1
else:
return 0
def board_to_plane(board):
"""Takes a board position and translates it into a plane of binary values"""
"""Old code that returned a vector"""
# 0-63,448-511: bishop on dark squares PNBRQK == 123456
# 64- 512-: black pawns, white pawns white = True, black = False
# 128- 576-: knight
# 192- 640-: bishop on light squares
# 256- 704-: rook
# 320- 768-: queen
# 384- 832-895: king
# The location of a square is described by:
# 448*(color white=1 black=0) + 64*(piecetype 1-6) + square
# - 64*3*is_a_bishop*is_on_light_square
# 896-961 : side to move color (white = 1s)
plane = np.zeros(896, dtype=int)
for square in range(64):
piece_type = board.piece_type_at(square)
if piece_type is None:
pass
else:
if board.color_at(square) == chess.WHITE:
piece_color = 1
else:
piece_color = 0
index = 448 * piece_color + 64 * (piece_type - 1) + square
if piece_type == chess.BISHOP:
# Bishops on white diagonals go in the 0 index instead of 3.
index = index - 64 * 3 * is_white_square(square)
plane[index] = 1
if board.turn == chess.WHITE:
col = np.ones(64)
else:
col = np.zeros(64)
# print(plane)
plane2 = np.concatenate((plane, col))
return plane2
def board_to_planev1(board):
"""Takes a board position and translates it into a planes of binary values"""
"""New code retains the 8x8 planes"""
# The resulting plane will be (8,8,15) - apparently tensorflow prefers channels last
# 0, 7: bishop on dark squares PNBRQK == 123456
# 1- 8-: black pawns, white pawns white = True, black = False
# 2- 9-: knight
# 3- 10-: bishop on light squares
# 4- 11-: rook
# 5- 12-: queen
# 6- 13 : king
# 14 : color white = 1
# The location of a square is described by:
# 448*(color white=1 black=0) + 64*(piecetype 1-6) + square
# - 64*3*is_a_bishop*is_on_light_square
# 896-961 : side to move color (white = 1s)
plane = np.zeros((8, 8, 15), dtype=int)
for square in range(64):
piece_type = board.piece_type_at(square)
if piece_type is None:
pass
else:
if board.color_at(square) == chess.WHITE:
piece_color = 1
else:
piece_color = 0
file = chess.square_file(square)
rank = chess.square_rank(square)
index = 7 * piece_color + piece_type -1
if piece_type == chess.BISHOP:
# Bishops on white diagonals go in the 0/7 index instead of 3/10.
index = index - 3 * is_white_square(square)
plane[rank, file, index] = 1
if board.turn == chess.WHITE:
col = 1
else:
col = 0
plane[:, :, 14] = col
return plane
def board_label(board):
"""Returns the training labels for the board from Syzygy lookup"""
# 0 draw for side-to-move, 1 win for side-to-move (more than 50 moves), 2 win for side-to-move
# -1 loss in more than 50, -2 loss in <50
with chess.syzygy.open_tablebase("./") as tablebase:
# board = chess.Board("8/2K5/4B3/3N4/8/8/4k3/8 b - - 0 1")
wdl = tablebase.probe_wdl(board)
# 0 draw, x win in x, -x loss in x
# counts may be off by 1
with chess.syzygy.open_tablebase("./") as tablebase:
# board = chess.Board("8/2K5/4B3/3N4/8/8/4k3/8 b - - 0 1")
dtz = tablebase.probe_dtz(board)
if wdl == 0:
win = 0
draw = 1
loss = 0
elif wdl > 0:
win = 1
draw = 0
loss = 0
else:
win = 0
draw = 0
loss = 1
if dtz > 0:
quality = 2000 - dtz
elif dtz < 0:
quality = -2000 - dtz
else:
quality = 0
return win, draw, loss, quality
def adjust_case(input_str):
"""This converts endgame descriptors so that the first block is capitalized"""
"""and the second block is lowercase. e.g. krpkq to KRPkq"""
lower = input_str.lower()
second_k = lower.find("k", 1)
# print(f"second k at {second_k}")
out1 = lower[:second_k].upper()
out2 = lower[second_k:]
output_str = out1+out2
if second_k == -1:
output_str = "fail"
return output_str
def ask_for_input():
need_input = True
while need_input:
balance = input("Please enter the endgame (e.g. KRkp): ")
balance = adjust_case(balance)
# print(f"After Adjusting: {balance}")
number = int(input("Please enter the number of training examples: "))
if number > 0:
need_input = False
if balance == "fail":
need_input = True
if need_input:
print("There was a problem with the inputs.")
return balance, number
def generate_training():
material_balance, target_count = ask_for_input()
plane_version = 'v1'
# Note: it took about 1:30 to generate 10,000 positions
X_train = []
y_train = []
for i in range(target_count):
my_board = gen_fen(material_balance)
my_plane = board_to_planev1(my_board)
my_label = board_label(my_board)
X_train.append(my_plane)
y_train.append(my_label)
print(i)
# Converts from a list to an array at the end; faster than concat array
X_train = np.stack(X_train, axis=0)
y_train = np.stack(y_train, axis=0)
print(X_train.shape)
print(y_train.shape)
print("frequency list:")
unique_elements, counts_elements = np.unique(y_train[:, 3], return_counts=True)
print("Frequency of unique values of the said array:")
print(np.asarray((unique_elements, counts_elements)))
outfile = "train_"+material_balance.lower()+str(int(target_count/1000))+"K"+plane_version+".npz"
# Save as a compressed npz file
np.savez_compressed(outfile, X_train=X_train, y_train=y_train)
print(f"Data saved to {outfile}")
# test that we can read the data
# print("Doing a test read of the data")
# npzfile2 = np.load(outfile)
# print("npz variables")
# print(npzfile.files)
# print(npzfile['y_train'])
# X_t2 = npzfile2['X_train']
# y_t2 = npzfile2['y_train']
# print(X_t2.shape)
# print(y_t2.shape)
#print(y_t2)
"""import_data.py"""
import tensorflow as tf
from matplotlib import pyplot as plt
import numpy as np
def import_endgame(filename):
npzfile2 = np.load(filename)
# print("npz variables")
# print(npzfile.files)
# print(npzfile['y_train'])
X_train = npzfile2['X_train']
y_train = npzfile2['y_train']
# Converting time to mate back to an number (0,x)
y_train[y_train[:, 3] < 0, 3] = (-2000 - y_train[y_train[:, 3] < 0, 3]) / 2
y_train[y_train[:, 3] > 0, 3] = (2001 - y_train[y_train[:, 3] > 0, 3]) / 2
y_train[:, 3] += 16
# Or, convert the int values into floats, this doesn't seem to matter
# X_train = X_train.astype(float)
# y_train = y_train.astype(float)
print("frequency list:")
unique_elements, counts_elements = np.unique(y_train[:, 3], return_counts=True)
print("Frequency of unique values of the said array:")
print(np.asarray((unique_elements, counts_elements)))
return X_train, y_train
"""define_model.py"""
def create_model_eg1(my_learning_rate):
"""Create and compile a deep neural net."""
# This is a first try to get a simple model that works
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(8, 8, 15)))
model.add(tf.keras.layers.Dense(units=32, activation='relu'))
model.add(tf.keras.layers.Dense(units=32, activation='relu'))
model.add(tf.keras.layers.Dense(units=1))
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=my_learning_rate),
loss="mean_squared_error",
metrics=['MeanSquaredError'])
return model
def create_model_eg5(my_learning_rate):
"""Create and compile a deep neural net."""
# This is a first try to get a simple model that works
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(8, 8, 15)))
model.add(tf.keras.layers.Dense(units=128, activation='relu'))
model.add(tf.keras.layers.Dense(units=128, activation='relu'))
model.add(tf.keras.layers.Dense(units=64, activation='relu'))
model.add(tf.keras.layers.Dense(units=1))
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=my_learning_rate),
loss="mean_squared_error",
metrics=['MeanSquaredError'])
return model
def create_model_eg_bin3orig(my_learning_rate):
"""Create and compile a deep neural net."""
# This is a first try to get a simple model that works
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), input_shape=(8,8,15), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(units=64, activation='relu'))
model.add(tf.keras.layers.Dense(units=33))
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=my_learning_rate),
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
def create_model_eg_bin3(my_learning_rate):
"""Create and compile a deep neural net."""
# This is a first try to get a simple model that works
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), input_shape=(8,8,15), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(units=64, activation='relu',
kernel_regularizer=tf.keras.regularizers.l2(l=0.1)))
model.add(tf.keras.layers.Dense(units=33))
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=my_learning_rate),
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
def create_model_eg_bin3b(my_learning_rate):
"""Create and compile a deep neural net."""
# L2 reg: 0.1 in both destroyed accuracy
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(
filters=256, kernel_size=(3,3), input_shape=(8,8,15), strides=(1, 1), padding='same',
kernel_regularizer=tf.keras.regularizers.l2(l=0.005)))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(
filters=128, kernel_size=(3,3), strides=(1, 1), padding='same'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(units=64, activation='relu',
kernel_regularizer=tf.keras.regularizers.l2(l=0.005)))
model.add(tf.keras.layers.Dense(units=33))
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=my_learning_rate),
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
"""train_model.py"""
def train_model(model, tb_callback, train_features, train_label, epochs,
batch_size=None, validation_split=0.1):
"""Train the model by feeding it data."""
history = model.fit(x=train_features, y=train_label, batch_size=batch_size,
epochs=epochs, shuffle=True, callbacks=[tb_callback],
validation_split=validation_split)
# To track the progression of training, gather a snapshot
# of the model's metrics at each epoch.
epochs = history.epoch
hist = pd.DataFrame(history.history)
return epochs, hist
"""main.py"""
def ask_gen_training():
yes_no = input("Do you need to generate endgame training data? ")
if len(yes_no) == 0:
pass
elif yes_no[0].lower() == "y":
generate_training()
def ask_train_net():
yes_no = input("Do you want to train a net? ")
if len(yes_no) == 0:
pass
elif yes_no[0].lower() == "n":
return
# Get the data to use for training
plane_version="v1"
material_balance, target_count = ask_for_input()
material_balance = material_balance.lower()
input_file = "train_"+material_balance+str(int(target_count/1000))+"K"+plane_version+".npz"
(x_train, y_train4) = import_endgame(input_file)
# Print a sample image
print(x_train.shape)
print(y_train4.shape)
print(y_train4[18])
y_train = y_train4[:, 3] # Column 3 is the score (-2000, 2000)
##############
# The following variables are the hyperparameters.
learning_rate = 0.003
epochs = 200
batch_size = 2000
validation_split = 0.2
# Establish the model's topography.
tb_callback = tf.keras.callbacks.TensorBoard(log_dir="logs/bin3/", histogram_freq=1)
my_model = create_model_eg_bin3b(learning_rate)
my_model.summary()
# Train the model on the normalized training set.
epochs, hist = train_model(my_model, tb_callback, x_train, y_train,
epochs, batch_size, validation_split)
# Plot a graph of the metric vs. epochs.
# list_of_metrics_to_plot = ['accuracy','val_accuracy']
list_of_metrics_to_plot = ['loss', 'val_loss']
plot_curve(epochs, hist, list_of_metrics_to_plot)
def set_options():
print(f"Using Tensorflow {tf.__version__}")
# The following lines adjust the granularity of reporting.
pd.options.display.max_rows = 10
pd.options.display.float_format = "{:.6f}".format # was .3f
# The following line improves formatting when ouputting NumPy arrays.
np.set_printoptions(linewidth=200)
if __name__ == '__main__':
set_options()
print("Welcome to chess_pos_gen")
print("We are going to generate a file of chess endgame training examples")
print("to use in training a neural net.")
ask_gen_training()
ask_train_net()
"""conv(128,128) dense(128,33)
conv(128,128) dense(64,33)
l2 regularization = 0.1 on dense(64) layer
l2 reg = 0.02 on 1 conv layer and the dense layer
"""