Hierarchical Tsetlin Machine in CUDA

Implements the Hierarchical Tsetlin Machine in CUDA.

Installation

pip install PyHierarchicalTsetlinMachineCUDA

or

git clone git@github.com:cair/PyHierarchicalTsetlinMachineCUDA.git
cd PyHierarchicalTsetlinMachineCUDA
python ./setup.py sdist
pip install dist/PyHierarchicalTsetlinMachineCUDA-0.2.2.tar.gz 

Examples

Noisy Parity Demo

Code: NoisyParityDemo.py

from PyHierarchicalTsetlinMachineCUDA.tm import TsetlinMachine
import numpy as np
from time import time
import PyHierarchicalTsetlinMachineCUDA.tm as tm

clauses = 32
s = 25.0
T = 250

train_data = np.loadtxt("./examples/NoisyParityTrainingData.txt").astype(np.uint32)
X_train = train_data[:,0:-1]
Y_train = train_data[:,-1]

test_data = np.loadtxt("./examples/NoisyParityTestingData.txt").astype(np.uint32)
X_test = test_data[:,0:-1]
Y_test = test_data[:,-1]

tm = TsetlinMachine(clauses, T, s, number_of_state_bits=8, boost_true_positive_feedback=0, hierarchy_structure=((tm.AND_GROUP, 3), (tm.OR_ALTERNATIVES, 3), (tm.AND_GROUP, 2), (tm.OR_ALTERNATIVES, 3), (tm.AND_GROUP, 2)))

print("\nAccuracy over 1000 epochs:\n")
for e in range(1000):
	start_training = time()
	tm.fit(X_train, Y_train, epochs=10, incremental=True)
	stop_training = time()

	start_testing = time()
	result = 100*(tm.predict(X_test) == Y_test).mean()
	stop_testing = time()

	tm.print_hierarchy()

	print("\n#%d Accuracy: %.2f%% Training: %.2fs Testing: %.2fs" % (e+1, result, stop_training-start_training, stop_testing-start_testing))

Output

python ./examples/NoisyParityData.py
python ./examples/NoisyParityDemo.py

Accuracy over 1000 epochs:

#1 Accuracy: 51.19% Training: 21.88s Testing: 1.37s
#2 Accuracy: 51.94% Training: 21.91s Testing: 1.37s
#3 Accuracy: 53.46% Training: 21.91s Testing: 1.37s
...
CLAUSE 1: (((((x0 ∧ x1) ∨ (x0 ∧ x1) ∨ (¬x0 ∧ ¬x1)) ∧ ((¬x0 ∧ ¬x1) ∨ (x0 ∧ x1) ∨ (x0 ∧ x1))) ∨ (((¬x0 ∧ ¬x1) ∨ (¬x0 ∧ ¬x1) ∨ (x0 ∧ x1)) ∧ ((x0 ∧ x1) ∨ (x0 ∧ x1) ∨ (¬x0 ∧ ¬x1))) ∨ (((x0 ∧ x1) ∨ (x0 ∧ x1) ∨ (¬x0 ∧ ¬x1)) ∧ (¬x0 ∨ (x0 ∧ x1) ∨ (¬x0 ∧ ¬x1)))) ∧ ((((x0 ∧ ¬x1) ∨ (x0 ∧ ¬x1) ∨ (x1 ∧ ¬x0)) ∧ ((x0 ∧ x1) ∨ (¬x0 ∧ ¬x1) ∨ (¬x0 ∧ ¬x1))) ∨ (((x1 ∧ ¬x0) ∨ (x0 ∧ ¬x1) ∨ (x0 ∧ ¬x1)) ∧ ((x0 ∧ x1) ∨ (¬x0 ∧ ¬x1) ∨ (x0 ∧ x1))) ∨ (((x1 ∧ ¬x0) ∨ (x1 ∧ ¬x0) ∨ (x0 ∧ ¬x1)) ∧ ((x0 ∧ x1) ∨ (¬x0 ∧ ¬x1) ∨ (¬x0 ∧ ¬x1)))))
...
#778 Accuracy: 99.71% Training: 32.57s Testing: 2.18s

Hex Demo

Source: https://en.wikipedia.org/wiki/Hex_(board_game)

Code: HexDemo.py

from PyHierarchicalTsetlinMachineCUDA.tm import TsetlinMachine
import numpy as np
from time import time
import PyHierarchicalTsetlinMachineCUDA.tm as tm
import argparse

def default_args(**kwargs):
    parser = argparse.ArgumentParser()
    parser.add_argument("--epochs", default=1000, type=int)
    parser.add_argument("--clauses", default=2000, type=int)
    parser.add_argument("--T", default=9000, type=int)
    parser.add_argument("--s", default=44.0, type=float)
    parser.add_argument("--q", default=1.0, type=float)
    parser.add_argument("--boost", default=1, type=int)
    parser.add_argument("--number_of_state_bits", default=7, type=int)
    parser.add_argument("--or_alternatives", default=60, type=int)
  
    args = parser.parse_args()
    for key, value in kwargs.items():
        if key in args.__dict__:
            setattr(args, key, value)
    return args

args = default_args()

data = np.loadtxt("./examples/hex_data.txt").astype(np.uint32)
X_train = data[:int(len(data)*0.8),0:-1]
Y_train = data[:int(len(data)*0.8),-1]

X_test = data[int(len(data)*0.8):,0:-1]
Y_test = data[int(len(data)*0.8):,-1]

tsetlin_machine = TsetlinMachine(args.clauses, args.T, args.s, weighted_clauses=False, number_of_state_bits=args.number_of_state_bits, boost_true_positive_feedback=args.boost, hierarchy_structure=((tm.AND_GROUP, 72), (tm.OR_ALTERNATIVES, args.or_alternatives), (tm.AND_GROUP, 4)))

print("\nAccuracy over %d epochs:\n" % (args.epochs))
for e in range(args.epochs):
	start_training = time()
	for b in range(10):
		tsetlin_machine.fit(X_train[b*len(Y_train)//10:(b+1)*len(Y_train)//10], Y_train[b*len(Y_train)//10:(b+1)*len(Y_train)//10], epochs=1, incremental=True)
	stop_training = time()

	start_testing = time()
	result_testing = 100*(tsetlin_machine.predict(X_test) == Y_test).mean()
	stop_testing = time()

	result_training = 100*(tsetlin_machine.predict(X_train) == Y_train).mean()

	print("#%d Training Accuracy: %.2f%% Testing Accuracy: %.2f%% Training Time: %.2fs Testing Time: %.2fs" % (e+1, result_training, result_testing, stop_training-start_training, stop_testing-start_testing))

Output

cd ./examples
make
./hex
cd ..
python3 ./examples/HexDemo.py

Accuracy over 1000 epochs:

#1 Training Accuracy: 97.98% Testing Accuracy: 97.46% Training Time: 37.02s Testing Time: 5.74s
#2 Training Accuracy: 99.19% Testing Accuracy: 98.70% Training Time: 34.44s Testing Time: 5.76s
#3 Training Accuracy: 99.49% Testing Accuracy: 98.92% Training Time: 33.60s Testing Time: 5.75s
#4 Training Accuracy: 99.65% Testing Accuracy: 99.05% Training Time: 33.15s Testing Time: 5.78s
#5 Training Accuracy: 99.73% Testing Accuracy: 99.10% Training Time: 32.86s Testing Time: 5.76s
...
#446 Training Accuracy: 100.00% Testing Accuracy: 99.45% Training Time: 32.09s Testing Time: 6.09s
#447 Training Accuracy: 100.00% Testing Accuracy: 99.46% Training Time: 32.08s Testing Time: 6.09s
#448 Training Accuracy: 100.00% Testing Accuracy: 99.45% Training Time: 32.10s Testing Time: 6.09s
#449 Training Accuracy: 100.00% Testing Accuracy: 99.47% Training Time: 32.06s Testing Time: 6.07s
#450 Training Accuracy: 100.00% Testing Accuracy: 99.45% Training Time: 31.96s Testing Time: 6.08s
...

Paper

A Tsetlin Machine for Logical Learning and Reasoning With AND-OR Hierarchies. Ole-Christoffer Granmo, et al., 2026. (Forthcoming)

Licence

MIT License

Copyright (c) 2026 Ole-Christoffer Granmo and the University of Agder

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.