def test_softmax_train_reshape_input(sample_train, sample_test):
Xtrain, ytrain = sample_train(count=300)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
softmax = Softmax()
loss = softmax.train(Xtrain, ytrain, reg=0, learning_rate=1e-6)
assert loss[0] > loss[-1]
def test_softmax_train_reshape_input(sample_train, sample_test):
Xtrain, ytrain = sample_train(count=300)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
softmax = Softmax()
loss = softmax.train(Xtrain, ytrain, reg=0, learning_rate=1e-6)
assert loss[0] > loss[-1]
def test_softmax_train(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
with pytest.raises(ValueError):
#Xtrain has the wrong shape (that is why there is a value error)
softmax = Softmax()
softmax.train(Xtrain, ytrain)
def test_softmax_train(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
with pytest.raises(ValueError):
#Xtrain has the wrong shape (that is why there is a value error)
softmax = Softmax()
softmax.train(Xtrain, ytrain)
def test_softmax_train_1(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
with pytest.raises(ValueError):
softmax = Softmax()
softmax.train(Xtrain, Xtrain)
def test_softmax_train_1(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
with pytest.raises(ValueError):
softmax = Softmax()
softmax.train(Xtrain, Xtrain)
def test_softmax_train_2(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
with pytest.raises(ValueError):
#this will catcha Valueerror associated with bad unpacking of a tuple
softmax = Softmax()
softmax.train(ytrain, ytrain)
def test_softmax_train_2(sample_train, sample_test):
#this test is designed to verify that input shapes are correct
Xtrain, ytrain = sample_train(count=40)
Xtest, ytest = sample_test(count=10)
Xtrain = np.reshape(Xtrain, (Xtrain.shape[0], -1))
Xtrain = np.hstack([Xtrain, np.ones((Xtrain.shape[0], 1))])
with pytest.raises(ValueError):
#this will catcha Valueerror associated with bad unpacking of a tuple
softmax = Softmax()
softmax.train(ytrain, ytrain)
def main():
X_train, y_train, X_val, y_val, X_test, y_test = gen_train_val_test(49000, 1000, 1000)
#run_softmax_naive(X_train, y_train)
softmax = Softmax()
tic = time.time()
softmax.train(X_train, y_train, learning_rate=2.782559e-06, reg=1e3,num_iters=3000,batch_size=200, verbose=True)
acc_train = evaluation(softmax, X_train, y_train)
acc_val = evaluation(softmax, X_val, y_val)
acc_test = evaluation(softmax, X_test, y_test)
print 'Train acc :{} Validation :{} Test :{}'.format(acc_train, acc_val, acc_test)
toc = time.time()
print 'That took %fs' % (toc - tic)
import h5py
from cs231n.classifiers import Softmax
from numpy import loadtxt
import numpy as np
h5f = h5py.File('img_data.h5','r')
X = h5f['dataset_1'][:]
h5f.close()
y = loadtxt("y_labels.txt", dtype=np.uint8, delimiter="\n", unpack=False)
#X_train = np.zeros((27116,196608))
#y_train = np.zeros(27116)
#X_val = np.zeros((5000,196608))
#y_val = np.zeros(5000)
X_train = X[8000:35117,:]
y_train = y[8000:35117]
X_val=X[3000:8000,:]
y_val=y[3000:8000]
# Generate a random softmax weight matrix and use it to compute the loss.
W = np.random.randn(196608, 5) * 0.0001
loss_vectorized, grad_vectorized = softmax_loss_vectorized(W, X_train, y_train, 0.00001)
softmax=Softmax()
loss_hist = softmax.train(X_train, y_train, learning_rate=1e-7, reg=5e4,
num_iters=1500, verbose=False)
y_train_pred = softmax.predict(X_train)
training_accuracy = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
val_accuracy = np.mean(y_val == y_val_pred)
print 'training accuracy: %f' % (np.mean(y_train == y_train_pred), )
print 'validation accuracy: %f' % (np.mean(y_val == y_val_pred), )
best_val = -1
best_softmax = None
# same from svm
learning_rates = [1e-7, 5e-8]
regularization_strengths = [5e4, 5e5]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
pass
import itertools
for lr, reg in itertools.product(learning_rates, regularization_strengths):
sm = Softmax()
sm.train(X_train, y_train, lr, reg, num_iters=1500)
y_pred = sm.predict(X_train)
train_accuracy = np.mean(y_pred == y_train)
y_pred = sm.predict(X_val)
val_accuracy = np.mean(y_pred == y_val)
results[(lr, reg)] = (train_accuracy, val_accuracy)
if val_accuracy > best_val:
best_val = val_accuracy
best_softmax = sm
################################################################################
# END OF YOUR CODE #
# Use the validation set to tune hyperparameters (regularization strength and
# learning rate). You should experiment with different ranges for the learning
# rates and regularization strengths; if you are careful you should be able to
# get a classification accuracy of over 0.35 on the validation set.
from cs231n.classifiers import Softmax
results = {}
best_val = -1
best_softmax = None
learning_rates = [1.3e-7]
regularization_strengths = [5e4]
for lr in learning_rates:
for reg in regularization_strengths:
softmax = Softmax()
softmax.train(X_train, y_train, lr, reg,
num_iters=500, verbose=True)
y_train_pred = softmax.predict(X_train)
y_train_acc = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
y_val_acc = np.mean(y_val == y_val_pred)
results[lr,reg] = (y_train_acc,y_val_acc)
cur_val = y_val_acc
if cur_val>best_val:
best_val = cur_val
best_softmax = softmax
# Print out results.
for lr, reg in sorted(results):
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [5e4, 1e8]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for lr in np.linspace(learning_rates[0], learning_rates[1], 5):
for rs in np.linspace(regularization_strengths[0],
regularization_strengths[1], 5):
softmax = Softmax()
loss_hist = softmax.train(X_train,
y_train,
learning_rate=lr,
reg=rs,
num_iters=1500,
verbose=True)
train_accuracy = np.mean(y_train == softmax.predict(X_train))
val_accuracy = np.mean(y_val == softmax.predict(X_val))
if best_val < val_accuracy:
best_val = val_accuracy
best_softmax = softmax
results[(lr, rs)] = train_accuracy, val_accuracy
################################################################################
loss_vectorized, grad_vectorized = softmax_loss_vectorized(
W, X_dev, y_dev, 0.000005)
toc = time.time()
print('vectorized loss: %e computed in %fs' % (loss_vectorized, toc - tic))
grad_difference = np.linalg.norm(grad_naive - grad_vectorized, ord='fro')
print('Loss difference: %f' % np.abs(loss_naive - loss_vectorized))
print('Gradient difference: %f' % grad_difference)
from cs231n.classifiers import Softmax
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7, 1e-6, 5e-6, 1e-5, 5e-5]
regularization_strengths = [1e4, 2e4, 2.5e4, 3e4, 4e4, 5e4]
softmax_model = Softmax()
import itertools
for learn, regul in itertools.product(learning_rates,
regularization_strengths):
loss_hist = softmax_model.train(X_train,
y_train,
learning_rate=learn,
reg=regul,
num_iters=150,
verbose=False)
y_train_pred = softmax_model.predict(X_train)
y_val_pred = softmax_model.predict(X_val)
train_accur = np.mean(y_train == y_train_pred)
val_accur = np.mean(y_val == y_val_pred)
# 利用验证集来微调超参数（正则化参数和学习率）
from cs231n.classifiers import Softmax
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7, 1.5e-7, 0.75e-7, 1.25e-7]
reg_strengths = [2.5e4, 3e4, 3.25e4, 3.5e4, 4e4, 4.5e4, 5e4]
# learning_rates = np.logspace(-10, 10, 10)
# reg_strengths = np.logspace(-3, 6, 10)
# 用验证集调整
for lr in learning_rates:
for rs in reg_strengths:
softmax = Softmax()
softmax.train(X_train,
y_train,
learning_rate=lr,
reg=rs,
num_iters=1000)
y_train_pred = softmax.predict(X_train)
acc_train = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
acc_val = np.mean(y_val == y_val_pred)
results[(lr, rs)] = (acc_train, acc_val)
if best_val < acc_val:
best_val = acc_val
best_softmax = None
learning_rates = [1e-7, 2e-7, 5e-7]
#regularization_strengths = [5e4, 1e8]
regularization_strengths = [(1 + 0.1 * i) * 1e4
for i in range(-3, 4)] + [(5 + 0.1 * i) * 1e4
for i in range(-3, 4)]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for lr in learning_rates:
for rs in regularization_strengths:
softmax = Softmax()
softmax.train(X_train, y_train, lr, rs, num_iters=2000)
y_train_pred = softmax.predict(X_train)
train_accuracy = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
val_accuracy = np.mean(y_val == y_val_pred)
if val_accuracy > best_val:
best_val = val_accuracy
best_softmax = softmax
results[(lr, rs)] = train_accuracy, val_accuracy
#pass
################################################################################
# END OF YOUR CODE #
################################################################################
# Print out results.
from cs231n.classifiers import Softmax
results = {}
best_val = -1
best_softmax = None
learning_rates = [3e-7, 1e-6, 3e-6]
regularization_strengths = [1e3, 3e3, 1e4]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for lr in learning_rates:
for reg in regularization_strengths:
softmax = Softmax()
softmax.train(X_train,
y_train,
learning_rate=lr,
reg=reg,
num_iters=1501,
verbose=False)
y_train_pred = softmax.predict(X_train)
training_accuracy = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
validation_accuracy = np.mean(y_val == y_val_pred)
results[(lr, reg)] = (training_accuracy, validation_accuracy)
if best_val < validation_accuracy:
best_val = validation_accuracy
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [2.5e4, 5e4]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
params = [(x, y) for x in learning_rates for y in regularization_strengths]
for lrate, regular in params:
softmax = Softmax()
loss_hist = softmax.train(X_train,
y_train,
learning_rate=lrate,
reg=regular,
num_iters=700,
verbose=True)
y_train_pred = softmax.predict(X_train)
accuracy_train = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
accuracy_val = np.mean(y_val == y_val_pred)
results[(lrate, regular)] = (accuracy_train, accuracy_val)
if (best_val < accuracy_val):
best_val = accuracy_val
best_softmax = softmax
################################################################################
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [5e4, 1e8]
################################################################################
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
num_iterations = 200
for lr in learning_rates:
for reg in regularization_strengths:
softmax = Softmax()
softmax.train(X_train,
y_train,
learning_rate=lr,
reg=reg,
num_iters=num_iterations)
y_train_pred = softmax.predict(X_train)
y_val_pred = softmax.predict(X_val)
train_accuracy = np.mean(y_train == y_train_pred)
valid_accuracy = np.mean(y_val == y_val_pred)
results[(lr, reg)] = (train_accuracy, valid_accuracy)
if valid_accuracy > best_val:
best_val = valid_accuracy
best_softmax = softmax
################################################################################
# END OF YOUR CODE #
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [5e4, 1e8]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
print '-------------_*****--------------'
print 'Tuning Parameters: '
for lr in np.arange(0.0000001, 0.0000005, 0.00000005):
for reg in np.arange(5e4, 1e8, 20000):
sm_iter = Softmax()
sm_iter.train(X_train,
y_train,
learning_rate=lr,
reg=reg,
num_iters=3000,
verbose=True)
y_val_pred_iter = sm_iter.predict(X_val)
y_train_pred_iter = sm_iter.predict(X_train)
val_acc = np.mean(y_val == y_val_pred_iter)
train_acc = np.mean(y_train == y_train_pred_iter)
results[(lr, reg)] = (train_acc, val_acc) # Turple Mapping
print 'Validation accuracy: %f' % (val_acc)
if (val_acc > best_val):
best_val = val_acc
best_softmax = sm_iter
learning_rates = [1e-7, 5e-7]
regularization_strengths = [2.5e4, 5e4]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
# *****START OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****
num_learning = len(learning_rates)
num_regularization = len(regularization_strengths)
for i in range(num_learning):
for j in range(num_regularization):
softmax = Softmax()
loss_hist = softmax.train(X_train, y_train, learning_rate=learning_rates[i], reg=regularization_strengths[j],
num_iters=1500, verbose=False)
y_train_pred = softmax.predict(X_train)
accuracy_train = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
accuracy_val = np.mean(y_val == y_val_pred)
results[(learning_rates[i], regularization_strengths[j])] = (accuracy_train, accuracy_val)
if accuracy_val > best_val:
best_val = accuracy_val
best_softmax = softmax
# pass
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [2.5e4, 5e4]
tic = time.clock()
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for ilr in np.arange(learning_rates[0], learning_rates[1],
0.05 * (learning_rates[1] - learning_rates[0])):
for ireg in np.arange(
regularization_strengths[0], regularization_strengths[1], 0.05 *
(regularization_strengths[1] - regularization_strengths[0])):
sm = Softmax()
_ = sm.train(X_train,
y_train,
learning_rate=ilr,
reg=ireg,
num_iters=1500,
verbose=True)
y_train_pred = sm.predict(X_train)
train_acc = np.mean(y_train == y_train_pred)
print('training accuracy: %f' % (train_acc))
y_val_pred = sm.predict(X_val)
val_acc = np.mean(y_val == y_val_pred)
if val_acc > best_val:
best_val = val_acc
best_softmax = sm
print('validation accuracy: %f' % (val_acc))
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7]
regularization_strengths = [6e4,4e4]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for lerRate in learning_rates:
for regStrength in regularization_strengths:
softmax = Softmax()
tic = time.time()
loss_hist = softmax.train(X_train, y_train, learning_rate=lerRate, reg=regStrength,
num_iters=1500, verbose=True)
toc = time.time()
print 'That took %fs' % (toc - tic)
plt.plot(loss_hist)
plt.xlabel('Iteration number')
plt.ylabel('Loss value')
plt.show()
y_train_pred = softmax.predict(X_train)
print 'training accuracy: %f' % (np.mean(y_train == y_train_pred), )
trainingAccu = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X_val)
print 'validation accuracy: %f' % (np.mean(y_val == y_val_pred), )
validationAccu = np.mean(y_val == y_val_pred)
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [2.5e4, 5e4]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
from copy import deepcopy
for lr in learning_rates:
for reg in regularization_strengths:
softmax=Softmax()
train_loss=softmax.train(X_train, y_train, learning_rate=lr, reg=reg, num_iters=3000,
batch_size=200, verbose=False)
y_train_pred=softmax.predict(X_train)
train_accuracy = np.mean(y_train==y_train_pred)
y_val_pred=softmax.predict(X_val)
val_accuracy = np.mean(y_val==y_val_pred)
results[(lr,reg)]=(train_accuracy,val_accuracy)
if val_accuracy>best_val:
best_val=val_accuracy
best_softmax=deepcopy(softmax)
################################################################################
# END OF YOUR CODE #
################################################################################
from cs231n.classifiers.softmax import softmax_loss_vectorized
tic = time.time()
loss_vectorized, grad_vectorized = softmax_loss_vectorized(W, X_dev, y_dev, 0.000005)
toc = time.time()
print('vectorized loss: %e computed in %fs' % (loss_vectorized, toc - tic))
# As we did for the SVM, we use the Frobenius norm to compare the two versions
# of the gradient.
grad_difference = np.linalg.norm(grad_naive - grad_vectorized, ord='fro')
print('Loss difference: %f' % np.abs(loss_naive - loss_vectorized))
print('Gradient difference: %f' % grad_difference)
###################################################################################
from cs231n.classifiers import Softmax, LinearSVM
svm = Softmax()
tic = time.time()
loss_hist = svm.train(X_train, y_train, learning_rate=1e-7, reg=2.5e4,
num_iters=1500, verbose=True)
toc = time.time()
print('That took %fs' % (toc - tic))
plt.plot(loss_hist)
plt.xlabel('Iteration number')
plt.ylabel('Loss value')
plt.show()
# Use the validation set to tune hyperparameters (regularization strength and
# learning rate). You should experiment with different ranges for the learning
# rates and regularization strengths; if you are careful you should be able to
# get a classification accuracy of over 0.35 on the validation set.
from cs231n.classifiers import Softmax
learning_rates = [1e-7, 5e-7, 1e-6, 5e-6]
regularization_strengths = [1e3, 5e3, 1e4, 5e4]
# results is dictionary mapping tuples of the form
# (learning_rate, regularization_strength) to tuples of the form
# (training_accuracy, validation_accuracy). The accuracy is simply the fraction
# of data points that are correctly classified.
results = {}
best_val = -1 # The highest validation accuracy that we have seen so far.
best_softmax = None # The LinearSVM object that achieved the highest validation rate.
for lr in learning_rates:
for reg in regularization_strengths:
sm = Softmax()
loss_hist = sm.train(X_train,
y_train,
learning_rate=lr,
reg=reg,
num_iters=1500,
verbose=False)
y_train_pred = sm.predict(X_train)
tr_acc = np.mean(y_train == y_train_pred)
#print('training accuracy: %f' % (tr_acc,))
y_val_pred = sm.predict(X_val)
val_acc = np.mean(y_val == y_val_pred)
#print('validation accuracy: %f' % (val_acc,))
results[(lr, reg)] = (tr_acc, val_acc)
if val_acc > best_val:
best_val = val_acc
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
# Your code
################################################################################
# END OF YOUR CODE #
################################################################################
lr_sample = np.linspace(learning_rates[0],learning_rates[1],1)
reg_sample = np.linspace(regularization_strengths[0],regularization_strengths[1],1)
tic = time.time()
for s1 in lr_sample:
for s2 in reg_sample:
soft = Softmax()
# getting the trained modal for the current svm object
soft.train(X_train,y_train,learning_rate = s1,reg = s2,num_iters = 1500,verbose=False)
y_train_predicted = soft.predict(X_train)
# validation of the current svm modal
train_accuracy = np.mean(y_train_predicted == y_train)
# validation of the trained modal on unknown set
y_val_predicted = soft.predict(X_val)
# validation of the current svm modal
val_accuracy = np.mean(y_val_predicted == y_val)
results[(s1,s2)] = (train_accuracy,val_accuracy)
if val_accuracy > best_val :
best_val = val_accuracy
best_softmax = soft
toc = time.time()
best_val = -1
best_softmax = None
learning_rates = [1e-8, 5e-7]
regularization_strengths = [5e-4, 1e-8]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for lr_idx in range(0,len(learning_rates)):
for reg_idx in range(0,len(regularization_strengths)):
lr_use = learning_rates[lr_idx]
reg_use = regularization_strengths[reg_idx]
sfmx = Softmax()
loss_hist = sfmx.train(X_dev, y_dev, learning_rate=lr_use, reg=reg_use,
num_iters=1500, verbose=True)
y_train_pred = sfmx.predict(X_dev)
y_val_pred = sfmx.predict(X_val)
acc_train = np.mean(y_train == y_train_pred)
acc_val = np.mean(y_val == y_val_pred)
if acc_val > best_val:
best_lr = lr_use
best_reg = reg_use
best_val = acc_val
best_sfmx = sfmx
results_once = {(lr_use, reg_use): (acc_train, acc_val)}
results.update(results_once)
################################################################################
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [5e4, 1e8]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
record=0
for i in learning_rates:
for j in regularization_strengths:
soft=Softmax()
soft.train(X_train,y_train,learning_rate=i,reg=j,num_iters=1500,verbose=True)
pred=soft.predict(X_val)
precise=np.mean((pred==y_val))
pred_train=soft.predict(X_train)
tmp=np.mean((pred_train==y_train))
results[i,j]=[tmp,precise]
if precise>best_val:
best_softmax=soft
best_val=precise
################################################################################
# END OF YOUR CODE #
################################################################################
# Print out results.
# 训练模型
X_train = np.reshape(X_train, (X_train.shape[0], -1)) # 1维展开
X_train = np.hstack([X_train, np.ones([X_train.shape[0], 1])])
X_test = np.reshape(X_test, (X_test.shape[0], -1)) # 1维展开
X_test = np.hstack([X_test, np.ones([X_test.shape[0], 1])])
num_class = 10
W = np.random.randn(X_train.shape[1], num_class) * 0.001
# 检查数值梯度和解析梯度
from cs231n.classifiers import softmax_loss_naive, softmax_loss_vectorized
loss, grad = softmax_loss_naive(W, X_train, y_train, 0.5)
from cs231n.gradient_check import grad_check_sparse
f = lambda w: softmax_loss_vectorized(w, X_train, y_train, 0.5)[0]
grad_check_sparse(f, W, grad)
from cs231n.classifiers import Softmax
classifer = Softmax()
loss_hist = classifer.train(X_train,
y_train,
verbose=True,
num_iters=5000,
batch_size=100)
plt.plot(loss_hist)
plt.xlabel('Step')
plt.ylabel('Loss')
plt.show()
# 泛化准确率
y_pred = classifer.predict(X_test)
accuracy = np.mean(y_pred == y_test)
print("Test accuracy:", accuracy)
# from cs231n.classifiers import KNearestNeighbor, KNN_test, KNN_train
X2_test_feats = np.dot(X_test_feats,
best_net.params['W1']) + best_net.params['b1']
learning_rates = [1e-5, 1e-4, 1e-3, 1e-2, 1e-1, 1e0]
regularization_strengths = [0, 1e1]
results = {}
best_val = -1
best_softmax = None
################################################################################
# Softmax
################################################################################
for _l in learning_rates:
for _r in regularization_strengths:
softmax = Softmax()
loss_hist = softmax.train(X2_train_feats,
y_train,
learning_rate=_l,
reg=_r,
num_iters=1500,
verbose=False)
y_train_pred = softmax.predict(X2_train_feats)
train_accuracy = np.mean(y_train == y_train_pred)
y_val_pred = softmax.predict(X2_val_feats)
val_accuracy = np.mean(y_val == y_val_pred)
print("train accuracy:{0}, val accuracy : {1}".format(
train_accuracy, val_accuracy))
from cs231n.classifiers import Softmax
results = {}
best_val = -1
best_softmax = None
learning_rates = [1e-7, 5e-7]
regularization_strengths = [5e4, 1e8]
################################################################################
# TODO: #
# Use the validation set to set the learning rate and regularization strength. #
# This should be identical to the validation that you did for the SVM; save #
# the best trained softmax classifer in best_softmax. #
################################################################################
for alpha in learning_rates:
for lam in regularization_strengths:
softmax = Softmax()
loss_hist = softmax.train(X_train,
y_train,
learning_rate=alpha,
reg=lam,
num_iters=500,
verbose=True)
y_tr_pred = softmax.predict(X_train)
training_accuracy = float(np.mean(y_tr_pred == y_train))
y_cv_pred = softmax.predict(X_val)
cv_accuracy = float(np.mean(y_cv_pred == y_val))
results[(alpha, lam)] = (training_accuracy, cv_accuracy)
#getting many NaN over here... probably bcs of learning rate/regularization,
# didn't seem to have bugs (as also runs correctly and accuracy is predicted well)
if cv_accuracy > best_val: