plt.show()
# Plot Training, testing ll vs iteration number
fig, ax = plt.subplots()
# Plot 0/1 loss
ax.plot(iter_train, train_01, lw=2, color="green", label=r"Train")
ax.plot(iter_train, test_01, lw=2, color="blue", label=r"Test")
# Plot BGD best fit on the testing set from before
plt.axhline(y=0.106, xmin=-1, xmax=100, ls="--", linewidth=2, color = 'k',
label="BGD Testing Best Fit")
# Format plot
ax.legend(loc="upper right")
ax.set_xlabel("Iteration")
ax.set_ylabel("0/1 Loss")
fig.tight_layout()
if save_plots:
fig.savefig("sgd_nn_mnist_multi_train_test_01.pdf")
plt.show()
#######################################################
#
# Compute, output final losses
#
#######################################################
val.summarize_loss(X_train, y_train_true, X_test, y_test_true, w, w0,
y_train_label=y_train, y_test_label=y_test,
classfn=cu.multi_logistic_classifier, lossfn=val.logloss_multi)
fig, ax = plt.subplots()
# Plot -(log likelikehood) to get logloss
ax.plot(iter_train, train_01, lw=2, color="green", label=r"Train")
ax.plot(iter_train, test_01, lw=2, color="blue", label=r"Test")
# Format plot
ax.legend(loc="upper right")
ax.set_xlabel("Iteration")
ax.set_ylabel("0/1 Loss")
fig.tight_layout()
if save_plots:
fig.savefig("bgd_mnist_multi_train_test_01.pdf")
plt.show()
#######################################################
#
# Compute, output final losses
#
#######################################################
val.summarize_loss(X_train,
y_train_true,
X_test,
y_test_true,
w,
w0,
y_train_label=y_train,
y_test_label=y_test,
classfn=cu.multi_logistic_classifier,
lossfn=val.logloss_multi)
# Format plot
ax.legend(loc="upper right")
ax.set_xlabel("Iteration")
ax.set_ylabel("LogLoss")
fig.tight_layout()
if save_plots:
fig.savefig("mnist_bin_train_test_ll.pdf")
plt.show()
# Plot Training, testing ll vs iteration number
fig, ax = plt.subplots()
# Plot -(log likelikehood) to get logloss
ax.plot(iter_train, train_01, lw=2, color="green", label=r"Train")
ax.plot(iter_train, test_01, lw=2, color="blue", label=r"Test")
# Format plot
ax.legend(loc="upper right")
ax.set_xlabel("Iteration")
ax.set_ylabel("0/1 Loss")
fig.tight_layout()
if save_plots:
fig.savefig("mnist_bin_train_test_01.pdf")
plt.show()
# Output loss metrics!
val.summarize_loss(X_train, y_train_true, X_test, y_test_true, w, w0,
classfn=cu.logistic_classifier, lossfn=val.logloss_bin)
if save_plots:
fig.savefig("mnist_bin_train_test_ll.pdf")
plt.show()
# Plot Training, testing ll vs iteration number
fig, ax = plt.subplots()
# Plot -(log likelikehood) to get logloss
ax.plot(iter_train, train_01, lw=2, color="green", label=r"Train")
ax.plot(iter_train, test_01, lw=2, color="blue", label=r"Test")
# Format plot
ax.legend(loc="upper right")
ax.set_xlabel("Iteration")
ax.set_ylabel("0/1 Loss")
fig.tight_layout()
if save_plots:
fig.savefig("mnist_bin_train_test_01.pdf")
plt.show()
# Output loss metrics!
val.summarize_loss(X_train,
y_train_true,
X_test,
y_test_true,
w,
w0,
classfn=cu.logistic_classifier,
lossfn=val.logloss_bin)