def exercise_1_1B():
print("exercise_1_1B")
X, y = load_data()
X = as3f.normalize_mnist_data(X)
X, y, X_train, y_train, X_test, y_test = as3f.randomize_and_split_data(X, y)
# Record the clf
best_gscv = None
best_sig = None
best_deg = None
for sig in [.01,.1,1,10,100]:
for deg in [1,2,3,4,5,6,7]:
svc_param = {"kernel":["precomputed"],
"C":[.1,1,10,100,1000,10000]}
G_train = compute_gram_X_Y(X_train,X_train,sig,deg)
print("\nsig ", sig, "deg",deg)
gscv = as3f.grid_search_SVC(G_train, y_train, SVC, 5, svc_param)
try:
if gscv.best_score_ > best_gscv.best_score_:
best_gscv = gscv
best_sig = sig
best_deg = deg
except:
best_gscv = gscv
best_sig = sig
best_deg = deg
print ("\nBest", str(round(abs(best_gscv.best_score_),5)), "sig:",best_sig,"deg",best_deg,"C",best_gscv.best_params_["C"])
def exercise_1_1C():
print("exercise_1_1C")
X, y = load_data()
X = as3f.normalize_mnist_data(X)
X, y, X_train, y_train, X_test, y_test = as3f.randomize_and_split_data(X, y)
print(X_train.shape, X_test.shape)
sig = 10
deg = 7
c = 0.1
G_train = compute_gram_X_Y(X_train,X_train,sig,deg)
clf = SVC(kernel="precomputed",C=c)
clf.fit(G_train,y_train)
G_test = compute_gram_X_Y(X_test,X_train)
y_pred = clf.predict(G_test)
print("C:",c,"Sig:",sig,"Deg:",deg)
print("Acc:",np.sum(y_pred == y_test)/len(y_test))
sig = .1
deg = 4
c = 10000
G_train = compute_gram_X_Y(X_train,X_train,sig,deg)
clf = SVC(kernel="precomputed",C=c)
clf.fit(G_train,y_train)
G_test = compute_gram_X_Y(X_test,X_train)
y_pred = clf.predict(G_test)
print("C:",c,"Sig:",sig,"Deg:",deg)
print("Acc:",np.sum(y_pred == y_test)/len(y_test))
def exercise_1():
print("A3 Ex1 and Ex2")
X, y = load_data()
X = as3f.normalize_mnist_data(X)
X, y, X_train, y_train, X_test, y_test = as3f.randomize_and_split_data(
X, y)
# Hypertuning values of C and gamma
test_params = [.1, 1, 10, 100, 1000]
# Setup each model
svc_params = [{
'kernel': ['linear'],
'C': test_params
}, {
'kernel': ['rbf'],
'C': test_params,
'gamma': test_params
}, {
'kernel': ['poly'],
'C': test_params,
'degree': [2, 3, 4, 5, 6]
}]
# Loop through each model
for svc_param in svc_params:
# CV and Regression using Grid Search Cross Validation
gscv = as3f.grid_search_SVC(X_train, y_train, SVC, 5, svc_param)
# Analysis and plot of each model
exercise_1_2(gscv, X_train, y_train, X_test, y_test)
def exercise_1_1vg():
print("exercise1_1vg")
X, y = load_data()
X = as3f.normalize_mnist_data(X)
X, y, X_train, y_train, X_test, y_test = as3f.randomize_and_split_data(
X, y)
# store best values
best_sig = None
best_deg = None
best_c = None
best_accuracy = None
for deg in [2]: #1,2,3,4,5]:
for c in [10000]: #1,10,100,1000,10000]:
for sig in [.01]: #.01,.1,1]:
# Compute gram X_train and X_train
G_train = compute_gram_X_Y(X_train, X_train, sig, deg)
# Create SVC instance and fit using precomputed gram matrix
clf = SVC(kernel="precomputed", C=c)
clf.fit(G_train, y_train)
# Predict using gram of X_test and X_train
G_test = compute_gram_X_Y(X_test, X_train)
y_pred = clf.predict(G_test)
# Get accuracy to find best hyperparameters
accuracy = np.sum(y_pred == y_test) / len(y_test)
#print("C:",c,"Sig:",sig,"Deg:",deg)
#print("Acc:", accuracy)
# store best hyperparameters
try:
if accuracy > best_accuracy:
best_sig = sig
best_deg = deg
best_c = c
best_accuracy = accuracy
except:
best_sig = sig
best_deg = deg
best_c = c
best_accuracy = accuracy
# print best hyperparameters
print("Best ", best_accuracy, "sig", best_sig, "deg", best_deg, "c",
best_c)
def exercise_1_1_1():
print("exercise1_1")
X, y = load_data()
X = as3f.normalize_mnist_data(X)
X, y, X_train, y_train, X_test, y_test = as3f.randomize_and_split_data(X, y, seed=55)
best_clf = None
best_sig = None
best_deg = None
best_c = None
best_accuracy = None
best_G_train = None
for deg in [2]:#1,2,3,4,5]:
for c in [10000]:#1,10,100,1000,10000]:
for sig in [.01]:#.01,.1,1]:
G_train = compute_gram_X_Y(X_train,X_train,sig,deg)
clf = SVC(kernel="precomputed",C=c)
clf.fit(G_train,y_train)
G_test = compute_gram_X_Y(X_test,X_train)
y_pred = clf.predict(G_test)
accuracy = np.sum(y_pred == y_test)/len(y_test)
print("C:",c,"Sig:",sig,"Deg:",deg)
print("Acc:", accuracy)
try:
if accuracy > best_accuracy:
best_clf = clf
best_sig = sig
best_deg = deg
best_c = c
best_accuracy = accuracy
best_G_train = G_train
except:
best_clf = clf
best_sig = sig
best_deg = deg
best_c = c
best_accuracy = accuracy
best_G_train = G_train
print ("Best ",best_accuracy, "sig",best_sig, "deg", best_deg, "c",best_c)