def run_grid_search(dataDicts, ada=False):
depths = [
1, 2, 3, 5, 10, 12, 13, 15, 16, 17, 18, 19, 20, 21, 23, 25, 27, 29, 30,
35
]
accuracies_beach = []
accuracies_finger = []
accuracies_rest = []
for depth in depths:
if ada == False:
clf_tree = tree.DecisionTreeClassifier(max_depth=depth,
random_state=0)
accuracy, _, _ = utils.cross_val(clf_tree, dataDicts)
accuracies_beach.append(accuracy["beach"])
accuracies_finger.append(accuracy["finger"])
accuracies_rest.append(accuracy["rest"])
else:
if depth > 15:
accuracies_beach.append(0)
accuracies_finger.append(0)
accuracies_rest.append(0)
else:
clf_tree = tree.DecisionTreeClassifier(max_depth=depth,
random_state=0)
accuracy, _, _ = utils.cross_val(clf_tree, dataDicts, ada=True)
accuracies_beach.append(accuracy["beach"])
accuracies_finger.append(accuracy["finger"])
accuracies_rest.append(accuracy["rest"])
keys = ["beach", "finger", "rest"]
accuracies = {
"beach": accuracies_beach,
"finger": accuracies_finger,
"rest": accuracies_rest,
"depths": depths
}
for key in keys:
data = accuracies[key]
ind = np.argmax(data)
print("dataset: " + key)
if ada == False:
print("optimal maximum depth value for Decision Tree: ")
print("depth value: " + str(depths[ind]) + ", accuracy: " +
str(data[ind]))
else:
print(
"optimal maximum depth value for Decision Tree with AdaBoost: "
)
print("depth value: " + str(depths[ind]) + ", accuracy: " +
str(data[ind]))
print("depths")
print(depths)
print("beach data")
print(accuracies_beach)
print("finger data")
print(accuracies_finger)
print("rest data")
print(accuracies_rest)
def compareCV(dataDict):
svm = SVC(kernel='linear')
acc_, _, _ = utils.cross_val(svm, dataDict)
print(acc_)
ada = AdaBoostClassifier(svm, algorithm='SAMME')
acc_, _, _ = utils.cross_val(ada, dataDict)
print(acc_)
def run(dataDicts, ada=False, max_depth=1):
if ada == False:
clf_tree = tree.DecisionTreeClassifier(max_depth=max_depth,
random_state=0)
accuracy, precision, recall = utils.cross_val(clf_tree, dataDicts)
else:
clf_tree = tree.DecisionTreeClassifier(max_depth=max_depth,
random_state=0)
accuracy, precision, recall = utils.cross_val(clf_tree,
dataDicts,
ada=True)
keys = ["beach", "finger", "rest"]
for key in keys:
print("dataset: " + key)
if ada == False:
print("Decision Tree")
print("Accuracy: " + str(accuracy[key]) + ", Recall: " +
str(recall[key]) + ", Precision: " + str(precision[key]))
else:
print("Decision Tree with AdaBoost")
print("Accuracy: " + str(accuracy[key]) + ", Recall: " +
str(recall[key]) + ", Precision: " + str(precision[key]))
def run(dataDicts):
clf = GaussianNB()
a, p, r = utils.cross_val(clf, dataDicts)
print(a)
def run(dataDicts):
clf = MLPClassifier(solver='lbfgs', alpha=0.01, random_state=1)
A, B, C = utils.cross_val(clf, dataDicts)
print(A)
# for a in Alpha:
# print(a)
# clf = MLPClassifier(solver='lbfgs', alpha=a, random_state=1)
# A, B, C = utils.cross_val(clf, dataDicts)
# print(A)
# for layer in HiddenLayerSize:
# print(layer)
# clf = MLPClassifier(solver='lbfgs', alpha=0.01, hidden_layer_sizes = (layer,), random_state=1)
# A, B, C = utils.cross_val(clf, dataDicts)
# print(A)
# for iter in MaxIter:
# print(iter)
# clf = MLPClassifier(solver='lbfgs', alpha=0.01, max_iter=iter, random_state=1)
# A, B, C = utils.cross_val(clf, dataDicts)
# print(A)
# for act in Activation:
# print(act)
# clf = MLPClassifier(solver='lbfgs', alpha = 0.01, activation = act, random_state=1)
# A, B, C = utils.cross_val(clf, dataDicts)
# print(A)
# for learn in LearningRate:
# print(learn)
# clf = MLPClassifier(solver='sgd', alpha=0.01, learning_rate_init = learn, random_state=1)
# A, B, C = utils.cross_val(clf, dataDicts)
# print(A)
warnings.filterwarnings('ignore')
for key in dataDicts[4]:
print(">> Dataset: " + key.upper())
[train_x, train_y, test_x, test_y] = dataDicts[4][key]
param_g = {
"alpha": [0.001, 0.01, 0.1, 1],
"hidden_layer_sizes": [(50, 50, 50), (100, 50), (100)],
"max_iter": [100, 250, 500, 1000],
"activation": ['identity', 'logistic', 'relu'],
"solver": ['adam'],
"random_state": [0, 1, 2, 3, 4],
"learning_rate_init": [0.001, 0.005, 0.01]
}
# inData = ""
# gs = RandomizedSearchCV(MLPClassifier(), param_distributions=param_g, scoring="accuracy", n_jobs=-1, n_iter=100)
# inData = "ARM"
# gs = GridSearchCV(MLPClassifier(random_state=1), param_grid=param_g, scoring="accuracy", n_jobs=-1)
# inData = "FARM"
# gs = GridSearchCV(MLPClassifier(random_state=1), param_grid=param_g, scoring="accuracy", n_jobs=-1)
# inData = "WARM"
# gs = GridSearchCV(MLPClassifier(random_state=1), param_grid=param_g, scoring="accuracy", n_jobs=-1)
# inData = "WARM" - test best random seed
gs = GridSearchCV(MLPClassifier(),
param_grid=param_g,
scoring="accuracy",
n_jobs=-1)
# inData = "NWARM"
# gs = GridSearchCV(MLPClassifier(random_state=1), param_grid=param_g, scoring="accuracy", n_jobs=-1)
gs.fit(train_x, train_y)
bestParam = gs.best_params_
print(bestParam)
bestScore = gs.best_score_
print(bestScore)
utils.printToTxt(modelName, "mlp_" + key + "_best", bestParam)
utils.printToTxt(modelName, "mlp_" + key + "_best_score", bestScore)
utils.printToTxt(modelName, "mlp_" + key + "_cv_results",
gs.cv_results_)