def demo(structure = [25**2, 23**2, 21**2,19**2,16**2, 15**2]):
# Getting the data
X,y = utils.load_mnist()
autoencoder = StackedDA([100], alpha=0.01)
autoencoder.pre_train(X[:1000], 10)
y = utils.makeMultiClass(y)
autoencoder.fine_tune(X[:1000], y[:1000], learning_layer=200, n_iters=20, alpha=0.01)
W = autoencoder.W[0].T[:, 1:]
W = utils.saveTiles(W, img_shape= (28,28), tile_shape=(10,10), filename="Results/res_dA.png")
def demo():
X,y = utils.load_mnist()
y = utils.makeMultiClass(y)
# building the SDA
sDA = StackedDA([100])
# pre-trainning the SDA
sDA.pre_train(X[:100], noise_rate=0.3, epochs=1)
# saving a PNG representation of the first layer
W = sDA.Layers[0].W.T[:, 1:]
utils.saveTiles(W, img_shape= (28,28), tile_shape=(10,10), filename="results/res_dA.png")
# adding the final layer
sDA.finalLayer(X[:37500], y[:37500], epochs=2)
# trainning the whole network
sDA.fine_tune(X[:37500], y[:37500], epochs=2)
# predicting using the SDA
pred = sDA.predict(X[37500:]).argmax(1)
# let's see how the network did
y = y[37500:].argmax(1)
e = 0.0
for i in range(len(y)):
e += y[i]==pred[i]
# printing the result, this structure should result in 80% accuracy
print "accuracy: %2.2f%%"%(100*e/len(y))
return sDA
def useLayers():
X, y = utils.load_mnist()
y = utils.makeMultiClass(y)
# Layers
sDA = StackedDA([100])
sDA.pre_train(X[:1000], rate=0.5, n_iters=500)
sDA.finalLayer(y[:1000], learner_size=200, n_iters=1)
sDA.fine_tune(X[:1000], y[:1000], n_iters=1)
pred = sDA.predict(X)
W = sDA.Layers[0].W.T[:, 1:]
W = utils.saveTiles(W,
img_shape=(28, 28),
tile_shape=(10, 10),
filename="Results/res_dA.png")
return pred, y
def useLayers():
X,y = utils.load_mnist()
y = utils.makeMultiClass(y)
# Layers
sDA = StackedDA([100])
sDA.pre_train(X[:1000], rate=0.5, n_iters=500)
sDA.finalLayer(y[:1000], learner_size=200, n_iters=1)
sDA.fine_tune(X[:1000], y[:1000], n_iters=1)
pred = sDA.predict(X)
W = sDA.Layers[0].W.T[:, 1:]
W = utils.saveTiles(W, img_shape= (28,28), tile_shape=(10,10), filename="Results/res_dA.png")
return pred, y
def demo(structure=[25**2, 23**2, 21**2, 19**2, 16**2, 15**2]):
# Getting the data
X, y = utils.load_mnist()
autoencoder = StackedDA([100], alpha=0.01)
autoencoder.pre_train(X[:1000], 10)
y = utils.makeMultiClass(y)
autoencoder.fine_tune(X[:1000],
y[:1000],
learning_layer=200,
n_iters=20,
alpha=0.01)
W = autoencoder.W[0].T[:, 1:]
W = utils.saveTiles(W,
img_shape=(28, 28),
tile_shape=(10, 10),
filename="Results/res_dA.png")
def SdA_gridsearch():
funcs = DLFuncs_SdA()
traindata_path='Z://Cristina//Section3//SdA_experiments//allLpatches.pklz'
trainUdata_path='Z://Cristina//Section3//SdA_experiments//allUpatches.pklz'
labeldata_path='Z://Cristina//Section3//SdA_experiments//allLabels.pklz'
datasets = funcs.load_wUdata(traindata_path, labeldata_path, trainUdata_path)
############
### plotting labels
############
dftrain = pd.DataFrame(); dfvalid = pd.DataFrame(); dftest = pd.DataFrame();
# get training data in numpy format
X,y = datasets[3]
Xtrain = np.asarray(X)
# extract one img
Xtrain = Xtrain.reshape(Xtrain.shape[0], 4, 900)
Xtrain = Xtrain[:,0,:]
ytrain = utils2.makeMultiClass(y)
dftrain['y'] = pd.Series(['y0' if y[yi]==0 else 'y1' for yi in range(len(ytrain))])
dftrain['group'] = pd.Series(np.repeat('train',len(y)))
# get valid data in numpy format
X,y = datasets[4]
Xvalid = np.asarray(X)
# extract one img
Xvalid = Xvalid.reshape(Xvalid.shape[0], 4, 900)
Xvalid = Xvalid[:,0,:]
yvalid = utils2.makeMultiClass(y)
dfvalid['y'] = pd.Series(['y0' if y[yi]==0 else 'y1' for yi in range(len(yvalid))])
dfvalid['group'] = pd.Series(np.repeat('valid',len(y)))
# get training data in numpy format
X,y = datasets[5]
Xtest = np.asarray(X)
# extract one img
Xtest = Xtest.reshape(Xtest.shape[0], 4, 900)
Xtest = Xtest[:,0,:]
ytest = utils2.makeMultiClass(y)
dftest['y'] = pd.Series(['y0' if y[yi]==0 else 'y1' for yi in range(len(ytest))])
dftest['group'] = pd.Series(np.repeat('test',len(y)))
# concat and plot
df = pd.concat([dftrain, dfvalid, dftest], axis=0)
sns.set(style="whitegrid")
# Draw a nested barplot to show survival for class and sex
g = sns.countplot(x="group", hue="y", palette="muted", data=df)
for p in g.patches:
height = p.get_height()
g.text(p.get_x(), height+ 3, '%1.2f'%(height))
fig = g.get_figure()
fig.savefig('grid_searchResults/datasets.png')
############
### Define grid search parameters
############
nlayers = [1,2,3,4]
nhiddens = [100,225,400,900]
hidden_layers_sidelen = [10,15,20,30]
nnoise_rate = [0.35]
nalpha = [0.01,0.1,0.5]
batch_size = [500,100,10]
# note batch_size is fixed to 1 and epochs is fixed to 25
k=0
BestAveAccuracy = 0
#dfresults = pd.DataFrame()# when first time
###########
## Process Resuts
###########
pkl_filegridS = open('grid_searchResults/gridSearch_results2.pkl','rb')
dfresults = pickle.load(pkl_filegridS)
print dfresults
#items = itertools.product(nlayers, nhiddens, nnoise_rate, nalpha, batch_size)
#item=items.next()
for item in itertools.product(nlayers, nhiddens, nnoise_rate, nalpha, batch_size):
k+=1
print(k,item)
if(k>90):
# setup the training functions
nlayer = item[0]
nhidden = item[1]
sidelen = hidden_layers_sidelen[ nhiddens.index(nhidden) ]
noiseRate = item[2]
alpha = item[3]
batchS = item[4]
if(nlayer == 1):
StackedDA_layers = [nhidden]
if(nlayer == 2):
StackedDA_layers = [nhidden,nhidden]
if(nlayer == 3):
StackedDA_layers = [nhidden,nhidden,nhidden]
# building the SDA
sDA = StackedDA(StackedDA_layers, alpha, item)
# pre-trainning the SDA
sDA.pre_train(Xtrain, noise_rate=noiseRate, epochs=200, batchsize=batchS)
#####################################
# saving a PNG representation of the first layer
#####################################
# Plot images in 2D
W0 = sDA.Layers[0].W.T[:, 1:]
imageW0 = Image.fromarray(
utils.tile_raster_images(X=W0 , img_shape=(30, 30),
tile_shape=(10, 10),
tile_spacing=(1, 1)))
#show and save
imageW0.save('grid_searchResults/filters_hidden_1stlayer_'+str(item)+'.png')
# prepare display
fig, ax = plt.subplots()
ax.imshow(imageW0, cmap="Greys_r")
ax.axes.get_xaxis().set_visible(False)
ax.axes.get_yaxis().set_visible(False)
# W = sDA.Layers[0].W.T[:, 1:]
#utils2.saveTiles(W, img_shape= (30,30), tile_shape=(20,10), filename='grid_searchResults/'+'.png")
# adding the final layer
sDA.finalLayer(Xtrain, ytrain, epochs=20)
# trainning the whole network
sDA.fine_tune(Xtrain, ytrain, epochs=20)
# predicting using the SDA
pred = sDA.predict(Xvalid).argmax(1)
# let's see how the network did
y = yvalid.argmax(1)
e0 = 0.0; y0 = len([0 for yi in range(len(y)) if y[yi]==0])
e1 = 0.0; y1 = len([1 for yi in range(len(y)) if y[yi]==1])
for i in range(len(y)):
if(y[i] == 1):
#print(y[i]==pred[i], y[i])
e1 += y[i]==pred[i]
if(y[i] == 0):
#print(y[i]==pred[i], y[i])
e0 += y[i]==pred[i]
# printing the result, this structure should result in 80% accuracy
print "accuracy for class 0: %2.2f%%"%(100*e0/y0)
print "accuracy for class 1: %2.2f%%"%(100*e1/y1)
# append results
accuracy0 = 100*e0/y0
accuracy1 = 100*e1/y1
itemresults = item + (accuracy0, accuracy1)
dSresultsiter = pd.DataFrame(data=np.array(itemresults)).T
dSresultsiter.columns=['nlayers', 'nhiddens', 'nnoise_rate', 'nalpha', 'batchsize', 'accuracy0','accuracy1']
dfresults = dfresults.append(dSresultsiter)
AveAccuracy = (accuracy0 + accuracy1)/2
# find best model so far
if( AveAccuracy > BestAveAccuracy):
bestsDA = sDA
BestAveAccuracy = AveAccuracy
print("best Accuracy = %d, for SdA:" % AveAccuracy)
print(bestsDA)
# save the best model
with open('grid_searchResults/gridSearch_results2.pkl', 'wb') as f:
pickle.dump(dfresults, f)
### continue
print(dfresults)
return
def main():
# open and load csv files
time_load_start = time.clock()
X_train, y_train = fipr.load_csv("train_file.csv", True)
X_test, y_test = fipr.load_csv("test_file.csv", True)
#y_train = y_train.flatten()
#y_test = y_test.flatten()
time_load_end = time.clock()
print("Loading finished, loading time: %g seconds" %
(time_load_end - time_load_start))
X_test_even, y_test_even = fipr.load_csv("test_file_even.csv", True)
training_data = X_train
training_labels = y_train
test_data = X_test
test_labels = y_test
test_data_even = X_test_even
test_labels_even = y_test_even
# building the SDA
sDA = StackedDA([100])
# start counting time for training
time_train_start = time.clock()
print('Pre-training...')
# pre-trainning the SDA
sDA.pre_train(training_data[:1000], noise_rate=0.3, epochs=100)
print('Training Network...')
# adding the final layer
sDA.finalLayer(training_data, training_labels, epochs=500)
# trainning the whole network
sDA.fine_tune(training_data, training_labels, epochs=500)
# print training time
time_train_end = time.clock()
print("Training finished, training time: %g seconds \n" %
(time_train_end - time_train_start))
# start counting time for testing
time_test_start = time.clock()
print('Testing performance...')
# predicting using the SDA
y_pred = sDA.predict(test_data).argmax(1)
# print simple precision metric to the console
print('Accuracy: ' + str(fipr.compute_accuracy(y_test, y_pred)))
# print testing time
time_test_end = time.clock()
print("Testing finished, testing time: %g seconds \n" %
(time_test_end - time_test_start))
# Even set test
y_pred_even = sDA.predict(test_data_even).argmax(1)
# print simple precision metric to the console
print('Accuracy on EVEN set: ' +
str(fipr.compute_accuracy(y_test_even, y_pred_even)))
return sDA
g = sns.countplot(x="group", hue="y", palette="muted", data=df)
a = ["1"]
b = ["0"]
c = ["a","b","c"]
d = ["d","e","f"]
for item in itertools.product(a, b, c, d):
print(item)
# building the SDA
sDA = StackedDA([400, 400])
# pre-trainning the SDA
sDA.pre_train(Xvalid, noise_rate=0.3, epochs=1)
# saving a PNG representation of the first layer
W = sDA.Layers[0].W.T[:, 1:]
utils2.saveTiles(W, img_shape= (30,30), tile_shape=(20,10), filename="res_dA.png")
# adding the final layer
sDA.finalLayer(Xtrain, ytrain, epochs=1)
# trainning the whole network
sDA.fine_tune(Xtrain, ytrain, epochs=1)
# predicting using the SDA