def createTree(X, y, minSampleSplit=2, maxDepth=None, depth=0):
X = np.array(X)
y = np.array(y)
tree = {}
depth = depth + 1
if ((isinstance(maxDepth, (int, float)) and depth <= maxDepth) or maxDepth
== None) and len(y) >= minSampleSplit and computeEntropy(y) != 0:
splits = computeSplits(X)
optimumSplitList = optimumSplit(X, y, splits)
columnIndex = list(optimumSplitList.keys())[0]
threshold = list(optimumSplitList.values())[0]
lowerX, upperX, lowerY, upperY = splitData(X, y, columnIndex,
threshold)
lowerNode = createTree(lowerX,
lowerY,
minSampleSplit=minSampleSplit,
depth=depth,
maxDepth=maxDepth)
upperNode = createTree(upperX,
upperY,
minSampleSplit=minSampleSplit,
depth=depth,
maxDepth=maxDepth)
tree[columnIndex] = [threshold, lowerNode, upperNode]
else:
return selectClass(y)
return tree
def optimumSplit(X, y, splits):
X = np.array(X)
y = np.array(y)
optimumSplittings = {}
minimumEntropy = 10000
optimumColumn = None
optimumThreshold = None
for columnIndex, thresholdList in splits.items():
for threshold in thresholdList:
x1, x2, y1, y2 = splitData(X, y, columnIndex, threshold)
if computeBranchEntropy(y1, y2) < minimumEntropy:
minimumEntropy = computeBranchEntropy(y1, y2)
optimumColumn = columnIndex
optimumThreshold = threshold
optimumSplittings[optimumColumn] = optimumThreshold
return optimumSplittings
data = loadmat('Data/Data.mat')
#X is a matrix containing Training Data
#Y is a matrix containing Training Labels
X = data['X']
y = data['y']
print('Displaying 100 Random Images')
rand_indices = np.random.permutation(range(X.shape[0]))
sel = X[rand_indices[0:100], :]
displayData(sel)
print('Seperating Data into Test and Training Sets')
print('\n')
#create Test and Train examples
X_test, X_train, y_train, Y_test, Y = splitData(X, y)
print('One Hot Encoding Labels')
print('\n')
encoder = OneHotEncoder(sparse=False, categories='auto')
y_onehot = encoder.fit_transform(y)
y_train = encoder.fit_transform(y_train)
print('Setting up Neural Network')
print('\n')
# initial setup
input_size = 400
hidden_size = 25
num_labels = 10
learning_rate = .9
reviews[i] = ' '.join(data[i][0:-1]) labels[i] = int(data[i][-1]) ##### convert labels with 0 to -1 ########### for i in range(len(labels)): if labels[i]==0: labels[i]=-1 labels = np.asarray(labels) path_to_weight_matrix = 'path to weight matrix' fname = 'weightmatrix.h5' weight_matrix_df = pd.read_hdf(fname) weight_matrix = weight_matrix_df.as_matrix() splits = 10 trainb_ilst,trainy_ilst,testb_ilst,testy_ilst = splitData(weight_matrix,labels,splits) split_idx = 0 #### determine train and test data ##### train_mat = np.array(trainb_ilst[split_idx]) train_labels = np.array(trainy_ilst[split_idx]) test_mat = np.array(testb_ilst[split_idx]) test_labels = np.array(testy_ilst[split_idx]) num_features = 80 #### set training parameters ########## # load guess matrix either with LSA or word2vec fname = 'lsaguessvectors.h5' word_guess_df = pd.read_hdf(fname) word_guess = word_guess_df.as_matrix()