test_y = mat["test_y"] # shape (20730, 1)
training_images = mat["training_images"] # shape (24, 24, 3, 21910)
training_y = mat["training_y"] # shape (21910, 1)
montage_n = 300
sort_ix = np.argsort(training_y, axis=0)
sort_ix_low = sort_ix[:montage_n] # get the 300 smallest
sort_ix_high = sort_ix[-montage_n:] #Get the 300 largest
# visualize the 300 smallest and the 300 largest nuclei
X_small = training_images[:, :, :, sort_ix_low.ravel()]
X_large = training_images[:, :, :, sort_ix_high.ravel()]
fig = plt.figure(figsize=(16, 8))
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
util.montageRGB(X_small, ax1)
ax1.set_title('300 smallest nuclei')
util.montageRGB(X_large, ax2)
ax2.set_title('300 largest nuclei')
# dataset preparation
imageSize = training_images.shape
# every pixel is a feature so the number of features is:
# height x width x color channels
numFeatures = imageSize[0] * imageSize[1] * imageSize[2]
training_x = training_images.reshape(numFeatures, imageSize[3]).T.astype(float)
test_x = test_images.reshape(numFeatures, test_images.shape[3]).T.astype(float)
## training linear regression model
def nuclei_measurement():
fn = '../data/nuclei_data.mat'
mat = scipy.io.loadmat(fn)
test_images = mat["test_images"] # shape (24, 24, 3, 20730)
test_y = mat["test_y"] # shape (20730, 1)
training_images = mat["training_images"] # shape (24, 24, 3, 21910)
training_y = mat["training_y"] # shape (21910, 1)
montage_n = 300
sort_ix = np.argsort(training_y, axis=0)
sort_ix_low = sort_ix[:montage_n] # get the 300 smallest
sort_ix_high = sort_ix[-montage_n:] #Get the 300 largest
# visualize the 300 smallest and the 300 largest nuclei
X_small = training_images[:, :, :, sort_ix_low.ravel()]
X_large = training_images[:, :, :, sort_ix_high.ravel()]
fig = plt.figure(figsize=(16, 8))
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
util.montageRGB(X_small, ax1)
ax1.set_title('300 smallest nuclei')
util.montageRGB(X_large, ax2)
ax2.set_title('300 largest nuclei')
# dataset preparation
imageSize = training_images.shape
# every pixel is a feature so the number of features is:
# height x width x color channels
numFeatures = imageSize[0] * imageSize[1] * imageSize[2]
training_x = training_images.reshape(numFeatures,
imageSize[3]).T.astype(float)
test_x = test_images.reshape(numFeatures,
test_images.shape[3]).T.astype(float)
## training linear regression model
#---------------------------------------------------------------------#
# TODO: Implement training of a linear regression model for measuring
# the area of nuclei in microscopy images. Then, use the trained model
# to predict the areas of the nuclei in the test dataset.
#---------------------------------------------------------------------#
# visualize the results
fig2 = plt.figure(figsize=(16, 8))
ax1 = fig2.add_subplot(121)
line1, = ax1.plot(test_y, predicted_y, ".g", markersize=3)
ax1.grid()
ax1.set_xlabel('Area')
ax1.set_ylabel('Predicted Area')
ax1.set_title('Training with full sample')
#training with smaller number of training samples
#---------------------------------------------------------------------#
# TODO: Train a model with reduced dataset size (e.g. every fourth
# training sample).
#---------------------------------------------------------------------#
# visualize the results
ax2 = fig2.add_subplot(122)
line2, = ax2.plot(test_y, predicted_y, ".g", markersize=3)
ax2.grid()
ax2.set_xlabel('Area')
ax2.set_ylabel('Predicted Area')
ax2.set_title('Training with smaller sample')
def nuclei_measurement(batch_size=1000):
fn = '../data/nuclei_data.mat'
mat = scipy.io.loadmat(fn)
test_images = mat["test_images"] # shape (24, 24, 3, 20730)
test_y = mat["test_y"] # shape (20730, 1)
training_images = mat["training_images"] # shape (24, 24, 3, 21910)
training_y = mat["training_y"] # shape (21910, 1)
montage_n = 300
sort_ix = np.argsort(training_y, axis=0)
sort_ix_low = sort_ix[:montage_n] # get the 300 smallest
sort_ix_high = sort_ix[-montage_n:] #Get the 300 largest
# visualize the 300 smallest and the 300 largest nuclei
X_small = training_images[:, :, :, sort_ix_low.ravel()]
X_large = training_images[:, :, :, sort_ix_high.ravel()]
fig = plt.figure(figsize=(16, 8))
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
util.montageRGB(X_small, ax1)
ax1.set_title('300 smallest nuclei')
util.montageRGB(X_large, ax2)
ax2.set_title('300 largest nuclei')
# dataset preparation
imageSize = training_images.shape
# every pixel is a feature so the number of features is:
# height x width x color channels
numFeatures = imageSize[0] * imageSize[1] * imageSize[2]
print((numFeatures))
training_x = training_images.reshape(numFeatures,
imageSize[3]).T.astype(float)
test_x = test_images.reshape(numFeatures,
test_images.shape[3]).T.astype(float)
#Predict Y (= area) for the test set and return the error too
E_test, predicted_y = sup.linear_regression(training_x, test_x,
numFeatures)
# visualize the results
fig2 = plt.figure(figsize=(16, 8))
ax1 = fig2.add_subplot(121)
line1, = ax1.plot(predicted_y, test_y, ".g", markersize=3)
ax1.grid()
ax1.set_xlabel('Area')
ax1.set_ylabel('Predicted Area')
ax1.set_title('Training with full sample')
#training with smaller number of training samples
#---------------------------------------------------------------------#
# TODO: Train a model with reduced dataset size (e.g. every fourth
# training sample).
#Choose the samples randomly, only set the number of samples (original 21910 samples)
ix = np.random.randint(imageSize[3], size=batch_size)
#Select the train data (only select certain samples)
training_x = training_images[:, :, :, ix].reshape(numFeatures,
len(ix)).T.astype(float)
E_test_small, predicted_y = sup.linear_regression(training_x, test_x,
batch_size)
#Evaluation
print(
"The error for testset using traindata consisting of all samples: {:.2f}"
.format(E_test))
print(
"The error for testset using traindata consisting of less samples: {:.2f}"
.format(E_test_small))
#---------------------------------------------------------------------#
# visualize the results
ax2 = fig2.add_subplot(122)
line2, = ax2.plot(predicted_y, test_y, ".g", markersize=3)
ax2.grid()
ax2.set_xlabel('Area')
ax2.set_ylabel('Predicted Area')
ax2.set_title('Training with smaller sample')
fig2.savefig(
"Predicted area and real area with batch size {}".format(batch_size))
return E_test, E_test_small