#Load Training Data
train_data = np.loadtxt("NoisyXORTrainingData.txt")
X_train = train_data[:, 0:-1] #last column is class labels
Y_train = train_data[:, -1] #last column is class labels
CLASSES = list(set(Y_train)) #list of classes
#Load Testing Data
test_data = np.loadtxt("NoisyXORTestData.txt")
X_test = test_data[:, 0:-1] #last column is class labels
Y_test = test_data[:, -1] #last column is class labels
#Initialize the Tsetlin Machine
tm = MultiClassTsetlinMachine(NUM_CLAUSES,
THRESHOLD,
S,
boost_true_positive_feedback=0)
#Fit TM on training data
tm.fit(X_train, Y_train, epochs=1)
#Predict on test data, compare to ground truth, calculate accuracy0
print("Accuracy:", 100 * (tm.predict(X_test) == Y_test).mean())
print('Type II feedbacks on clauses: ', tm.typeII_feedback_clauses)
#Prediction on some random data
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" %
feature_count_negated_negative = np.zeros(NUM_FEATURES)
clauses = np.zeros((RUNS * NUM_CLAUSES, NUM_FEATURES * 2 + 1))
clause_dict = {}
for r in range(RUNS):
print('Run:', r)
x_train, x_test, y_train, y_test = train_test_split(data, labels)
x_train_ids = x_train[:, -1]
x_test_ids = x_test[:, -1]
x_train = x_train[:, :-1]
x_test = x_test[:, :-1]
#print('\nsplits ready:',x_train.shape, x_test.shape)
tm = MultiClassTsetlinMachine(NUM_CLAUSES, T, s)
tm.fit(x_train, y_train, epochs=TRAIN_EPOCHS, incremental=True)
print('\nfit done')
result[r] = 100 * (tm.predict(x_test) == y_test).mean()
feature_vector = np.zeros(NUM_FEATURES * 2)
for cur_cls in CLASSES:
for cur_clause in range(NUM_CLAUSES):
if cur_clause % 2 == 0:
clause_type = 'positive'
else:
clause_type = 'negative'
this_clause = ''
for f in range(0, NUM_FEATURES):
action_plain = tm.ta_action(int(cur_cls), cur_clause, f)
action_negated = tm.ta_action(int(cur_cls), cur_clause,
f + NUM_FEATURES)
train_data = np.loadtxt("NoisyXORTrainingData.txt")
X_train = train_data[:, 0:-1] #last column is class labels
Y_train = train_data[:, -1] #last column is class labels
CLASSES = list(set(Y_train)) #list of classes
#Load Testing Data
test_data = np.loadtxt("NoisyXORTestData.txt")
X_test = test_data[:, 0:-1] #last column is class labels
Y_test = test_data[:, -1] #last column is class labels
print(X_train.shape)
print(Y_train.shape)
'''
#Initialize the Tsetlin Machine
tm = MultiClassTsetlinMachine(NUM_CLAUSES, THRESHOLD, S, boost_true_positive_feedback=0)
#Fit TM on training data
tm.fit(X_train, Y_train, epochs=1)
##save
tm.save_model('tm_model.npz', Y_train)
print("Accuracy:", 100*(tm.predict(X_test) == Y_test).mean())
'''
tm2 = MultiClassTsetlinMachine.load_model('tm_model.npz')
#Predict on test data, compare to ground truth, calculate accuracy0
#
print("Accuracy after saving:", 100 * (tm2.predict(X_test) == Y_test).mean())
Xtest = binarize_selected(combo_test, list_of_uniques, cs)
print('binarized train', Xtrain.shape)
print('binarized test', Xtest.shape)
#X_train = Xtrain.reshape((Xtrain.shape[0],len(addendum_context),1,usum))
#X_test = Xtest.reshape((Xtest.shape[0],len(addendum_context),1,usum))
#print('reshaped train',X_train.shape)
#print('reshaped test',X_test.shape)
#np.save('x_train_conv', Xtrain)
#np.save('x_test_conv', Xtest)
# Setup
tm = MultiClassTsetlinMachine(CLAUSES, T, s, weighted_clauses=weighting)
labels_test_indx = np.where(labels_test == 1)
labels_train_indx = np.where(labels_train == 1)
acc = []
acc_train = []
# Training
for i in range(RUNS):
print(i)
start_training = time()
tm.fit(Xtrain, labels_train, epochs=50, incremental=True)
stop_training = time()
start_testing = time()
res_test = tm.predict(Xtest)
res_train = tm.predict(Xtrain)
print(reverse_word_map)
print(word_idx)
print(sents[10], data[10])
x_train, x_test, y_train, y_test = train_test_split(data, labels)
x_train_ids = x_train[:, -1]
x_test_ids = x_test[:, -1]
x_train = x_train[:, :-1]
x_test = x_test[:, :-1]
NUM_CLAUSES = 20
T = 15
s = 3.9
print('\nsplits ready:', x_train.shape, x_test.shape)
tm = MultiClassTsetlinMachine(NUM_CLAUSES, T, s)
tm.fit(x_train, y_train, epochs=200, incremental=True)
print('\nfit done')
result = 100 * (tm.predict(x_test) == y_test).mean()
print(result)
res = tm.predict(x_test)
for i in range(len(x_test_ids)):
sidx = x_test_ids[i]
print(sents[sidx], res[i])
NUM_CLAUSES = 10
NUM_FEATURES = len(x_train[0])
CLASSES = list(set(y_train))
print('Num Clauses:', NUM_CLAUSES)
s=3#s=10 weighting = True training_epoch=5 RUNS=100 ''' CLAUSES=250 T=60 s=37 weighting = True training_epoch=5 RUNS=100 X_train, X_test, y_train, y_test = train_test_split(featureset_transformed_X, featureset_transformed_y, test_size=0.30, random_state=42, shuffle=True) tm = MultiClassTsetlinMachine(CLAUSES, T, s, weighted_clauses=weighting,append_negated=False) allacc=[] for i in range(RUNS): tm.fit(X_train, y_train, epochs=training_epoch, incremental=True) res_test=tm.predict(X_test) print(res_test) acc_test = 100*(res_test == y_test).mean() allacc.append(acc_test) #print(type(y_test), type(res_test), len(y_test), len(res_test)) prf_test_macro=precision_recall_fscore_support(list(res_test), list(y_test), average='macro') prf_test_macro=[str(round(p,2)) for p in prf_test_macro[:-1]]
X_train = training[:, 0:2]
Y_train = training[:, -1]
X_test = test[:, 0:2]
Y_test = test[:, -1]
CLASSES = list(set(Y_train)) #list of classes
NUM_FEATURES = len(X_train[0]) #number of features
print('Num Clauses:', NUM_CLAUSES)
print('Num Classes: ', len(CLASSES), ' : ', CLASSES)
print('Num Features: ', NUM_FEATURES)
tm = MultiClassTsetlinMachine(NUM_CLAUSES,
THRESHOLD,
S,
boost_true_positive_feedback=0)
tm.fit(X_train, Y_train, epochs=200)
print("Accuracy:", 100 * (tm.predict(X_test) == Y_test))
all_clauses = [[] for i in range(NUM_CLAUSES)]
for cur_cls in CLASSES:
for cur_clause in range(NUM_CLAUSES):
this_clause = ''
for f in range(NUM_FEATURES * 2):
action = tm.ta_action(int(cur_cls), cur_clause, f)
if action == 1:
if this_clause != '':
y = df2.iloc[:, :].values
y = np.reshape(y, len(y))
x_train, x_test, y_train, y_test = train_test_split(x,
y,
test_size=0.084,
shuffle=False)
print(x_train.shape)
print(x_test.shape)
print(y_train.shape)
print(y_test.shape)
print('data fitting started..........................')
tm = MultiClassTsetlinMachine(2000, 40, 27, weighted_clauses=False)
print("\nAccuracy over 1000 epochs:\n")
tempAcc = []
for i in range(500):
start_training = time()
tm.fit(x_train, y_train, epochs=1, incremental=True)
stop_training = time()
start_testing = time()
result1 = 100 * (tm.predict(x_test) == y_test).mean()
result2 = 100 * (tm.predict(x_train) == y_train).mean()
stop_testing = time()
tempAcc.append(result1)
print(
"#%d AccuracyTrain: %.2f%% AccuracyTest: %.2f%% Training: %.2fs Testing: %.2fs"
X_final1 = np.concatenate((X_text, X_target), axis=1)
X_final2 = np.concatenate((Loc_vec1, Loc_vec2), axis=1)
X_final3 = np.concatenate((X_final1, X_final2), axis=1)
X_final4 = np.concatenate((X_final3, restSC), axis=1)
#Split training and testing samples
X_train = X_final4[0:
3608, :] #For Laptop dataset change this from 3608 to 2328
X_test = X_final4[3608:, :] #For Laptop dataset change this from 3608 to 2328
ytrain = y[0:3608] #For Laptop dataset change this from 3608 to 2328
ytest = y[3608:] #For Laptop dataset change this from 3608 to 2328
#%%%%%%%%%%%%%%%%%% Initialize Tsetlin Machine %%%%%%%%%%%%%%%%%%%%%%%
tm1 = MultiClassTsetlinMachine(
700, 90 * 100, 15,
weighted_clauses=True) #number of clause= 700, T = 90*100 and s = 15
#tm1.fit(X_train, ytrain, epochs=0)
print("\nTraining Classification Layer...")
print("\nAccuracy over 1000 epochs:\n")
max = 0
for i in range(500):
start_training = time()
tm1.fit(X_train, ytrain, epochs=1, incremental=True)
stop_training = time()
start_testing = time()
result2 = 100 * (tm1.predict(X_train) == ytrain).mean()
result1 = 100 * (tm1.predict(X_test) == ytest).mean()
y_pred = tm1.predict(X_test)
CLAUSES=33
T=28
s=12.5
weighting = True
training_epoch=5
RUNS=100
featureset=np.load(fname.replace('.txt','')+'_featureset.npy')
X=featureset[:,:-1]
y=featureset[:,-1]
y=[int(yy) for yy in y]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30, random_state=42, shuffle=False)
tm = MultiClassTsetlinMachine(CLAUSES, T, s, weighted_clauses=weighting,append_negated=False)
allacc=[]
for i in range(RUNS):
tm.fit(X_train, y_train, epochs=training_epoch, incremental=True)
res_test=tm.predict(X_test)
acc_test = 100*(res_test == y_test).mean()
allacc.append(acc_test)
prf_test_macro=precision_recall_fscore_support(res_test, y_test, average='macro')
prf_test_macro=[str(round(p,2)) for p in prf_test_macro[:-1]]
prf_test_micro=precision_recall_fscore_support(res_test, y_test, average='micro')
prf_test_micro=[str(round(p,2)) for p in prf_test_micro[:-1]]
data = np.loadtxt("NoisyXORTestData.txt")
np.random.shuffle(data)
training, test = data[:4000, :], data[4000:, :]
X_train = training[:, 0:2]
Y_train = training[:, -1]
X_test = test[:, 0:2]
Y_test = test[:, -1]
CLASSES = list(set(Y_train)) #list of classes
NUM_FEATURES = len(X_train[0]) #number of features
tm = MultiClassTsetlinMachine(NUM_CLAUSES,
THRESHOLD,
S,
boost_true_positive_feedback=0)
#Fit TM on training data
tm.fit(X_train, Y_train, epochs=1)
'''
print('Train: ',len(X_train))
print('\nTest: ', len(X_test))
print('\nNum Clauses:', NUM_CLAUSES)
print('\nNum Classes: ', len(CLASSES),' : ', CLASSES)
print('\nNum Features: ', NUM_FEATURES)
print("Accuracy:", 100*(tm.predict(X_test) == Y_test).mean())
for cur_cls in CLASSES:
for cur_clause in range(NUM_CLAUSES):
feature_count_ignore = np.zeros(NUM_FEATURES)
feature_count_contradiction = np.zeros(NUM_FEATURES)
feature_count_negated_negative= np.zeros(NUM_FEATURES)
clauses=np.zeros((RUNS*NUM_CLAUSES,NUM_FEATURES*2+1))
clause_dict={}
for r in range(RUNS):
print('Run:',r)
x_train, x_test, y_train, y_test = train_test_split(data, labels)
x_train_ids=x_train[:,-1]
x_test_ids=x_test[:,-1]
x_train=x_train[:,:-1]
x_test=x_test[:,:-1]
tm = MultiClassTsetlinMachine(NUM_CLAUSES, T, s)
tm.fit(x_train, y_train, epochs=TRAIN_EPOCHS, incremental=True)
print('\nfit done')
tm.save_model('causal_model.npz')
res=tm.predict(x_test)
print('\n\nFull result average=',100*(res == y_test).mean())
for i in range(len(x_test_ids)):
sidx=x_test_ids[i]
print(sents[sidx], '\nPredicted:',res[i],'\nActual',y_test[i])
tm.predict_and_print(np.array([x_test[i]]))
result[r] = 100*(tm.predict(x_test) == y_test).mean()
feature_vector=np.zeros(NUM_FEATURES*2)
for cur_cls in CLASSES:
for cur_clause in range(NUM_CLAUSES):
if cur_clause%2==0:
fo.write('\ns: '+str(s))
fo.write('\nNum Features: '+ str(NUM_FEATURES))
fo.write('\nTotal Runsper Step: '+str(RUNS))
fo.write('\nTotal Steps: '+str(STEPS))
fo.write('\nSTEP Size: '+str(STEP_SIZE))
fo.write('\nTrain Epochs: '+str(TRAIN_EPOCHS)+'\n\n')
fo.write('Num_CLAUSES\tMean\tMax\tAll\n')
result_mean=np.zeros(STEPS)
result_max=np.zeros(STEPS)
clausesizes=np.zeros(STEPS)
for s in range(STEPS):
lr=np.zeros(RUNS)
NUM_CLAUSES+=STEP_SIZE
tm = MultiClassTsetlinMachine(NUM_CLAUSES, T, s)
for r in range(RUNS):
x_train, x_test, y_train, y_test = train_test_split(data, labels)
x_train_ids=x_train[:,-1]
x_test_ids=x_test[:,-1]
x_train=x_train[:,:-1]
x_test=x_test[:,:-1]
print ('Step '+str(s)+' Run '+str(r)+' num_clause '+str(NUM_CLAUSES))
tm.fit(x_train, y_train, epochs=TRAIN_EPOCHS, incremental=True)
lr[r]=100*(tm.predict(x_test) == y_test).mean()
result_mean[s] = lr.mean()
result_max[s] = lr.max()
clausesizes[s]=NUM_CLAUSES
fo.write(str(NUM_CLAUSES)+'\t'+str(result_mean[s])+'\t'+str(result_max[s])+'\t'+str(lr)+'\n')