Update README.md
Browse files
README.md
CHANGED
@@ -86,39 +86,41 @@ def preprocess(example, useful_fn):
|
|
86 |
<br>
|
87 |
<br>
|
88 |
<br>
|
|
|
89 |
# Complete Example
|
90 |
|
91 |
```py
|
92 |
import random
|
93 |
-
import
|
|
|
94 |
|
95 |
-
#
|
96 |
-
dataset = dataset.shuffle(seed=42)
|
97 |
-
dataset = dataset.map(preprocess, fn_kwargs={'tokenizer': tokenizer})
|
98 |
-
```
|
99 |
-
|
100 |
-
For a quick working example, you can try to use the following:
|
101 |
-
```py
|
102 |
-
# A mock tokenizer and preprocess function for demonstration.
|
103 |
class Tokenizer:
|
|
|
|
|
104 |
def __call__(self, example):
|
105 |
return example
|
106 |
|
107 |
-
def
|
|
|
|
|
|
|
|
|
108 |
# Get number of moves made in the game.
|
109 |
-
|
110 |
-
|
111 |
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
|
117 |
-
# Transform data into the format of your choice.
|
118 |
-
example['fens'] = useful_fn(fen)
|
119 |
-
example['moves'] = useful_fn(move)
|
120 |
-
example['scores'] = useful_fn(score)
|
121 |
-
return example
|
122 |
|
123 |
tokenizer = Tokenizer()
|
124 |
|
@@ -128,11 +130,23 @@ dataset = load_dataset(path="mauricett/lichess_sf",
|
|
128 |
streaming=True,
|
129 |
trust_remote_code=True)
|
130 |
|
|
|
131 |
# Shuffle and apply your own preprocessing.
|
132 |
dataset = dataset.shuffle(seed=42)
|
133 |
-
dataset = dataset.map(preprocess, fn_kwargs={'tokenizer': tokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
-
|
136 |
-
|
137 |
-
|
|
|
138 |
```
|
|
|
86 |
<br>
|
87 |
<br>
|
88 |
<br>
|
89 |
+
|
90 |
# Complete Example
|
91 |
|
92 |
```py
|
93 |
import random
|
94 |
+
from datasets import load_dataset
|
95 |
+
from torch.utils.data import DataLoader
|
96 |
|
97 |
+
# A mock tokenizer and functions for demonstration.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
class Tokenizer:
|
99 |
+
def __init__(self):
|
100 |
+
pass
|
101 |
def __call__(self, example):
|
102 |
return example
|
103 |
|
104 |
+
def score_fn(score):
|
105 |
+
# Transform Stockfish score and terminal outcomes.
|
106 |
+
return score
|
107 |
+
|
108 |
+
def preprocess(example, tokenizer, score_fn):
|
109 |
# Get number of moves made in the game.
|
110 |
+
max_ply = len(example['moves'])
|
111 |
+
pick_random_move = random.randint(0, max_ply-1)
|
112 |
|
113 |
+
# Get the FEN, move and score for our random choice.
|
114 |
+
fen = example['fens'][pick_random_move]
|
115 |
+
move = example['moves'][pick_random_move]
|
116 |
+
score = example['scores'][pick_random_move]
|
117 |
+
|
118 |
+
# Transform data into the format of your choice.
|
119 |
+
example['fens'] = tokenizer(fen)
|
120 |
+
example['moves'] = tokenizer(move)
|
121 |
+
example['scores'] = score_fn(score)
|
122 |
+
return example
|
123 |
|
|
|
|
|
|
|
|
|
|
|
124 |
|
125 |
tokenizer = Tokenizer()
|
126 |
|
|
|
130 |
streaming=True,
|
131 |
trust_remote_code=True)
|
132 |
|
133 |
+
|
134 |
# Shuffle and apply your own preprocessing.
|
135 |
dataset = dataset.shuffle(seed=42)
|
136 |
+
dataset = dataset.map(preprocess, fn_kwargs={'tokenizer': tokenizer,
|
137 |
+
'score_fn': score_fn})
|
138 |
+
|
139 |
+
# PyTorch dataloader
|
140 |
+
dataloader = DataLoader(dataset, batch_size=256, num_workers=4)
|
141 |
+
|
142 |
+
|
143 |
+
n = 0
|
144 |
+
for batch in dataloader:
|
145 |
+
|
146 |
+
# do stuff
|
147 |
|
148 |
+
n += 1
|
149 |
+
print(n)
|
150 |
+
if n == 50:
|
151 |
+
break
|
152 |
```
|