XOR example from PyTsetlinMachineCUDA does not train

[ricpruss]

I am probably doing something wrong but XOR example from PyTsetlinMachineCUDA does not seem to train with this model.

from pathlib import Path

import numpy as np
from PyCoalescedTsetlinMachineCUDA.tm import (
    MultiClassTsetlinMachine as MultiClassTsetlinMachine2,
)

train_data = np.loadtxt(
    Path.home() / "github/PyTsetlinMachineCUDA/examples/NoisyXORTrainingData.txt"
)

test_data = np.loadtxt(
    Path.home() / "github/PyTsetlinMachineCUDA/examples/NoisyXORTestData.txt"
)
X_test = test_data[:, 0:-1]
Y_test = test_data[:, -1]


tm = MultiClassTsetlinMachine2(10, 15, 3.9, boost_true_positive_feedback=0)

print("Starting training...")
for e in range(200):
    for i in range(100, len(train_data), 100):
        X_train = train_data[i - 100 : i, 0:-1]
        Y_train = train_data[i - 100 : i, -1]
        tm.fit(X_train, Y_train, epochs=1, incremental=True)
    print("Epoch:", e, " Accuracy:", 100 * (tm.predict(X_test) == Y_test).mean())

print(
    "Prediction: x1 = 1, x2 = 0, ... -> y = %d"
    % (tm.predict(np.array([[1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0]])))
)
print(
    "Prediction: x1 = 0, x2 = 1, ... -> y = %d"
    % (tm.predict(np.array([[0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0]])))
)
print(
    "Prediction: x1 = 0, x2 = 0, ... -> y = %d"
    % (tm.predict(np.array([[0, 0, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0]])))
)
print(
    "Prediction: x1 = 1, x2 = 1, ... -> y = %d"
    % (tm.predict(np.array([[1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0]])))
)

The results seem sad:
Starting training...
Epoch: 0 Accuracy: 48.8
Epoch: 1 Accuracy: 49.04
Epoch: 2 Accuracy: 48.980000000000004
Epoch: 3 Accuracy: 49.04
Epoch: 4 Accuracy: 48.64
Epoch: 5 Accuracy: 48.78
Epoch: 6 Accuracy: 48.82
Epoch: 7 Accuracy: 42.66
Epoch: 8 Accuracy: 48.88

[olegranmo]

Hi @ricpruss! Our Coalesced TM implementation has clause weighting turned on by default while the original XOR demo assumes standard clauses without weighting (default in the other implementations). For the Coalesced TM implementations it is currently not possible to turn the weighting off, however, we are working on addressing this.