def runSoftmaxOnMNISTMod3():
"""
-Runs softmaxRegression on MNIST training set with new trainYMod3
-computes test error by running computeTestErrorMod3()
-plots cost function over number of iters
-save new model to file: thetaM.pkl.gz
"""
print("[INFO] Running soft max on MNISTMod3")
trainX, trainY, testX, testY = getMNISTData()
trainYMod3, testYMod3 = updateY(trainY, testY)
print("[INFO] Loaded in data (mod3).")
print("[INFO] Using tempParameter of:", tempParameter)
start = time.time()
theta, costFunctionHistory = softmaxRegression(trainX,
trainYMod3,
tempParameter,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
finish = time.time()
print("[INFO] Finished training in %f secs" % (finish - start))
plotCostFunctionOverTime(costFunctionHistory)
testErrorMod3 = computeTestErrorMod3(testX, testYMod3, theta,
tempParameter)
# Save the model parameters theta obtained from calling softmaxRegression to disk.
writePickleData(theta, "./thetaM.pkl.gz")
return testErrorMod3
def runSoftmaxOnMNIST():
print("[INFO] Running soft max on MNIST")
trainX, trainY, testX, testY = getMNISTData()
print("[INFO] Loaded in data.")
print("[INFO] Using tempParameter of:", tempParameter)
start = time.time()
theta, costFunctionHistory = softmaxRegression(trainX,
trainY,
tempParameter,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
finish = time.time()
print("[INFO] Finished training in %f secs" % (finish - start))
#plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(testX, testY, theta, tempParameter)
print("Test Error:", testError)
# Save the model parameters theta obtained from calling softmaxRegression to disk.
writePickleData(theta, "./theta.pkl.gz")
# compute error mod 3
trainYMod3, testYMod3 = updateY(trainY, testY)
testErrorMod3 = computeTestErrorMod3(testX, testYMod3, theta,
tempParameter)
print("Test Error (mod 3):", testErrorMod3)
return (testError, testErrorMod3)
def runSoftmaxOnMNIST():
trainX, trainY, testX, testY = getMNISTData()
theta, costFunctionHistory = softmaxRegression(trainX, trainY, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(testX, testY, theta)
writePickleData(theta, "./theta.pkl.gz") # Save the model parameters theta obtained from calling softmaxRegression to disk.
return testError
def runSoftmaxOnMNISTMod3():
trainX, trainY, testX, testY = getMNISTData()
(trainYMod3, testYMod3) = updateY(trainY, testY)
theta, costFunctionHistory = softmaxRegression(trainX, trainY, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
# plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestErrorMod3(testX, testYMod3, theta, tempParameter)
writePickleData(theta, "./thetaM.pkl.gz")
return testError
def runSoftmaxOnMNISTMod3(): #YOUR CODE HERE trainX, trainY, testX, testY = getMNISTData() trainY_mod3 , testY_mod3 = updateY(trainY, testY) theta, costFunctionHistory = softmaxRegression(trainX, trainY_mod3, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150) plotCostFunctionOverTime(costFunctionHistory) testError = computeTestError(testX, testY_mod3, theta, tempParameter) # Save the model parameters theta obtained from calling softmaxRegression to disk. writePickleData(theta, "./thetaM.pkl.gz") return testError
def runSoftmaxQF():
theta, costFunctionHistory = softmaxRegression(train_quad,
trainY,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(test_quad, testY, theta)
return testError
def runSoftmaxOnMNIST_pca():
trainX, trainY, testX, testY = getMNISTData()
theta, costFunctionHistory = softmaxRegression(train_pca, trainY, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
# plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(test_pca, testY, theta, tempParameter)
# Save the model parameters theta obtained from calling softmaxRegression to disk.
# writePickleData(theta, "./theta.pkl.gz")
# TODO: add your code here for the "Changing labels" section (7)
# and print the testErrorMod3
return testError
def runSoftmaxOnMNISTPCA():
trainX, trainY, testX, testY = getMNISTData()
theta, costFunctionHistory = softmaxRegression(train_pca,
trainY,
tempParameter,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(test_pca, testY, theta, tempParameter)
return testError
def runSoftmaxOnMNIST():
trainX, trainY, testX, testY = getMNISTData()
theta, costFunctionHistory = softmaxRegression(trainX,
trainY,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(testX, testY, theta)
writePickleData(
theta, "./theta.pkl.gz"
) # Save the model parameters theta obtained from calling softmaxRegression to disk.
return testError
def runSoftmaxOnMNIST_PCA_Cubic():
"""
print("[INFO] Running soft max on MNIST")
trainX, trainY, testX, testY = getMNISTData()
print("[INFO] Loaded in data.")
print("[INFO] Using tempParameter of:", tempParameter)
start = time.time()
theta, costFunctionHistory = softmaxRegression(trainX, trainY, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
finish = time.time()
print("[INFO] Finished training in %f secs" % (finish-start))
#plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(testX, testY, theta, tempParameter)
print("Test Error:", testError)
"""
print(
"[INFO] Running soft max on MNIST data with PCA 10 and Cubic Kernel..."
)
tempParameter = 1.0
print("[INFO] Using tempParameter of:", tempParameter)
### LOAD DATA ###
trainX, trainY, testX, testY = getMNISTData()
print("[INFO] Loaded in data...")
### Find the 10-dimensional PCA representation of the training and test set ###
n_components = 10
pcs = principalComponents(trainX)
train_pca10 = projectOntoPC(trainX, pcs, n_components)
test_pca10 = projectOntoPC(testX, pcs, n_components)
print("[INFO] Got PCA features...")
### CONVERT TO CUBIC REPRESENTATION ###
train_cube = cubicFeatures(train_pca10)
test_cube = cubicFeatures(test_pca10)
print("[INFO] Got cubic features...")
### TRAIN THE MODEL ###
start = time.time()
theta, costFunctionHistory = softmaxRegression(train_cube,
trainY,
tempParameter,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=150)
finish = time.time()
print("[INFO] Finished training in %f secs" % (finish - start))
### PLOTTING ###
plotCostFunctionOverTime(costFunctionHistory)
### COMPUTE ERROR ###
err = computeTestError(test_cube, testY, theta, tempParameter)
print(" ### TEST ERROR RESULTS ###")
print("Test Error (PCA10 w/ Cubic Features):", err)
return err
#err = runSoftmaxOnMNIST_PCA_Cubic()
def runSoftmaxOnMNIST_PCA():
"""
print("[INFO] Running soft max on MNIST")
trainX, trainY, testX, testY = getMNISTData()
print("[INFO] Loaded in data.")
print("[INFO] Using tempParameter of:", tempParameter)
start = time.time()
theta, costFunctionHistory = softmaxRegression(trainX, trainY, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
finish = time.time()
print("[INFO] Finished training in %f secs" % (finish-start))
#plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(testX, testY, theta, tempParameter)
print("Test Error:", testError)
"""
tempParameter = 1.0
print("[INFO] Running soft max on MNIST data with PCA")
print("[INFO] Using tempParameter of:", tempParameter)
trainX, trainY, testX, testY = getMNISTData()
print("[INFO] Loaded in data...")
n_components = 18
pcs = principalComponents(trainX)
train_pca = projectOntoPC(trainX, pcs, n_components)
test_pca = projectOntoPC(testX, pcs, n_components)
print(test_pca)
print("[INFO] Projected data into PCA feature dimension...")
# # TODO: Train your softmax regression model using (train_pca, trainY)
# # and evaluate its accuracy on (test_pca, testY).
start = time.time()
print("[INFO] Dims of train_pca and trainY:", np.shape(train_pca),
np.shape(trainY))
print("[INFO] Dims of test_pca and testY:", np.shape(test_pca),
np.shape(testY))
theta, costFunctionHistory = softmaxRegression(train_pca,
trainY,
tempParameter,
alpha=0.3,
lambdaFactor=1.0e-4,
k=10,
numIterations=40)
finish = time.time() # 0.9244
print("[INFO] Finished training in %f secs" % (finish - start))
# plot and compute error
plotCostFunctionOverTime(costFunctionHistory)
print("Temp Param:", tempParameter)
testError = computeTestError(test_pca, testY, theta, tempParameter)
print("Test Error (PCA):", testError)
#TODO: Use the plotPC function in features.py to produce scatterplot
#of the first 100 MNIST images, as represented in the space spanned by the
#first 2 principal components found above.
plotPC(trainX[range(100), ], pcs, trainY[range(100)])
#TODO: Use the reconstructPC function in features.py to show
#the first and second MNIST images as reconstructed solely from
#their 18-dimensional principal component representation.
#Compare the reconstructed images with the originals.
firstimage_reconstructed = reconstructPC(train_pca[0, ], pcs, n_components,
trainX)
plotImages(firstimage_reconstructed)
plotImages(trainX[0, ])
secondimage_reconstructed = reconstructPC(train_pca[1, ], pcs,
n_components, trainX)
plotImages(secondimage_reconstructed)
plotImages(trainX[1, ])
### Dimensionality reduction via PCA ##
#
## TODO: First fill out the PCA functions in features.py as the below code depends on them.
n_components = 18
pcs = principalComponents(trainX)
train_pca = projectOntoPC(trainX, pcs, n_components)
test_pca = projectOntoPC(testX, pcs, n_components)
#train_pca (and test_pca) is a representation of our training (and test) data
#after projecting each example onto the first 18 principal components.
tempParameter = 1
## TODO: Train your softmax regression model using (train_pca, trainY)
## and evaluate its accuracy on (test_pca, testY).
trainX, trainY, testX, testY = getMNISTData()
theta, costFunctionHistory = softmaxRegression(train_pca, trainY, tempParameter, alpha= 0.3, lambdaFactor = 1.0e-4, k = 10, numIterations = 150)
plotCostFunctionOverTime(costFunctionHistory)
testError = computeTestError(test_pca, testY, theta, tempParameter)
print(' testError =', testError)
#
#
## TODO: Use the plotPC function in features.py to produce scatterplot
## of the first 100 MNIST images, as represented in the space spanned by the
### first 2 principal components found above.
plotPC(trainX[range(100),], pcs, trainY[range(100)])
#
#
## TODO: Use the reconstructPC function in features.py to show
## the first and second MNIST images as reconstructed solely from
## their 18-dimensional principal component representation.
## Compare the reconstructed images with the originals.
