def som(self, m, n, dim, iterations):
# step 4
trainData = numpy.zeros((len(self.contrastPoints), 1))
for i in range(len(self.contrastPoints)):
trainData[i] = [self.contrastPoints[i]]
som = SOM(m, n, dim, iterations)
som.train(trainData)
mapped = som.map_vects(trainData)
d0 = 0
d1 = 0
d2 = 0
for d, m in zip(trainData, mapped):
mapv = m[0] + m[1]
if (d0 == 0 and mapv == 0):
d0 = d[0] if d0 < d[0] else d0
if (d1 == 0 and mapv == 1):
d1 = d[0] if d1 < d[0] else d1
if (d2 == 0 and mapv == 2):
d2 = d[0] if d2 < d[0] else d2
least = 0 if d0 < d1 and d0 < d2 else (1 if d1 < d0 and d1 < d2 else (2 if d2 < d1 and d2 < d0 else 0))
most = 0 if d0 > d1 and d0 > d2 else (1 if d1 > d0 and d1 > d2 else (2 if d2 > d1 and d2 > d0 else 0))
mean = 0 if d0 > d1 and d0 < d2 else (1 if d1 > d0 and d1 < d2 else (2 if d2 > d1 and d2 < d0 else 0))
print(d0, d1, d2)
print(least, most, mean)
self.mappingDict = {'least': least, 'mean': mean, 'most': most}
self.blocksMapping = mapped
def SelfOrganizingMap(Feature_List):
som = SOM(SOM_LINES, SOM_COLS, len(Feature_List[0]), SOM_ITERATIONS)
som.train(Feature_List)
image_grid = som.get_centroids()
mapped = som.map_vects(Feature_List)
return mapped, image_grid
def call_som(m, n, dim, input_data, weight_after_insertion=None, alpha=None, sigma=None):
# get the shape of input data inorder to calc iternumber in SOM
iter_no = input_data.shape[0]*10
som = SOM(m, n, dim, weight_after_insertion, n_iterations= iter_no)
trained_weight = som.train(input_data)
mapped = som.map_vects(input_data)
result = np.array(mapped)
return trained_weight, result
def SelfOrganizingMap(feature_list):
no_features = len(feature_list[0])
print('Clustering', no_features, 'features')
som = SOM(SOM_LINES, SOM_COLS, no_features, SOM_ITERATIONS, alpha=ALPHA, sigma=SIGMA)
som.train(feature_list)
image_grid = som.get_centroids()
mapped = som.map_vects(feature_list)
return mapped, image_grid
def call_som(m,
n,
dim,
input_data,
weight_after_insertion=None,
alpha=None,
sigma=None):
som = SOM(m, n, dim, weight_after_insertion)
trained_weight = som.train(input_data)
mapped = som.map_vects(input_data)
result = np.array(mapped)
return trained_weight, result
def main():
# Training inputs for RGBcolors
credits_data = pd.read_csv("testdata/credit_train.csv", header=None)
columns = credits_data.columns
train_data = credits_data[columns[:-1]].copy().values
label_data = credits_data[columns[-1]].copy().values
train_data_scale = preprocessing.scale(train_data)
# colors = np.array(
# [[0., 0., 0.],
# [0., 0., 1.],
# [0., 0., 0.5],
# [0.125, 0.529, 1.0],
# [0.33, 0.4, 0.67],
# [0.6, 0.5, 1.0],
# [0., 1., 0.],
# [1., 0., 0.],
# [0., 1., 1.],
# [1., 0., 1.],
# [1., 1., 0.],
# [1., 1., 1.],
# [.33, .33, .33],
# [.5, .5, .5],
# [.66, .66, .66]])
# color_names = \
# ['black', 'blue', 'darkblue', 'skyblue',
# 'greyblue', 'lilac', 'green', 'red',
# 'cyan', 'violet', 'yellow', 'white',
# 'darkgrey', 'mediumgrey', 'lightgrey']
# Train a 20x30 SOM with 400 iterations
som = SOM(1, 2, len(columns) - 1, 400)
som.train(train_data_scale)
# Get output grid
image_grid = som.get_centroids()
# Map colours to their closest neurons
mapped = som.map_vects(train_data_scale)
# Plot
plt.imshow(image_grid)
plt.title('Color SOM')
for i, m in enumerate(mapped):
plt.text(m[1],
m[0],
label_data[i],
ha='center',
va='center',
bbox=dict(facecolor='white', alpha=0.5, lw=0))
plt.show()
def SortingSelfOrganizingMap(Feature_List):
# Train a SS_LINES x SS_COLS self-organizing map using NUM_ITERATIONS iterations
som = SOM(SS_LINES, SS_COLS, len(Feature_List[0]), NUM_ITERATIONS)
som.train(Feature_List)
mapped = som.map_vects(Feature_List)
# Grayscale landscape
sorting_landscape = np.zeros((SS_LINES, SS_COLS), np.float16)
for i in range(len(mapped)):
lin = mapped[i][0]
col = mapped[i][1]
sorting_landscape[lin][col] = Feature_List[i][0]
io.imsave('sorted_landscape.png', sorting_landscape)
def som3(self, m, n, dim, iterations):
# step 4
trainData = numpy.zeros((len(self.contrastPoints), 3))
for i in range(len(self.contrastPoints)):
trainData[i] = [self.diagonalContrast[i], self.verticalContrast[i], self.horizontalContrast[i]]
som = SOM(m, n, dim, iterations)
som.train(trainData)
mapped = som.map_vects(trainData)
d0 = 0
d1 = 0
d2 = 0
for d, m in zip(trainData, mapped):
mapv = m[0] + m[1]
if (d0 == 0 and mapv == 0):
d0 = d[0] if d0 < d[0] else d0
if (d1 == 0 and mapv == 1):
d1 = d[0] if d1 < d[0] else d1
if (d2 == 0 and mapv == 2):
d2 = d[0] if d2 < d[0] else d2
arr = numpy.array([d0, d1, d2])
most = numpy.argmax(arr)
numpy.delete(arr, most, 0)
least = numpy.argmin(arr)
numpy.delete(arr, least, 0)
mean = 0 if (arr[most] == d1 or arr[most] == d2) and (arr[least] == d1 or arr[least] == d2) else (1 if (arr[most] == d0 or arr[most] == d2) and (arr[least] == d0 or arr[least] == d2) else 2)
self.mappingDict = {'least': least, 'mean': mean, 'most': most}
self.blocksMapping = mapped
self.neuronsMap = som.get_centroids()
n_iter = 100
som = SOM(m, n, 3, n_iter)
for iter_no in range(n_iter):
#Train with each vector one by one
for i in range(len(data)):
som(data[i], iter_no)
#Store a centroid grid for easy retrieval later on
centroid_grid = [[] for i in range(m)]
weights = som.get_weights()
locations = som.get_locations()
for i, loc in enumerate(locations):
centroid_grid[loc[0]].append(weights[i].numpy())
#Get output grid
image_grid = centroid_grid
#Map colours to their closest neurons
mapped = som.map_vects(torch.Tensor(colors))
#Plot
plt.imshow(image_grid)
plt.title('Color SOM')
for i, m in enumerate(mapped):
plt.text(m[1],
m[0],
color_names[i],
ha='center',
va='center',
bbox=dict(facecolor='white', alpha=0.5, lw=0))
plt.show()
n_iter = 100
som = SOM(m, n, 3, n_iter)
for iter_no in range(n_iter):
#Train with each vector one by one
for i in range(len(data)):
som(data[i], iter_no)
#Store a centroid grid for easy retrieval later on
centroid_grid = [[] for i in range(m)]
weights = som.get_weights()
locations = som.get_locations()
for i, loc in enumerate(locations):
centroid_grid[loc[0]].append(weights[i].numpy())
#Get output grid
image_grid = centroid_grid
#Map colours to their closest neurons
mapped = som.map_vects(colors)
#Plot
plt.imshow(image_grid)
plt.title('Color SOM')
for i, m in enumerate(mapped):
plt.text(m[1],
m[0],
color_names[i],
ha='center',
va='center',
bbox=dict(facecolor='white', alpha=0.5, lw=0))
plt.show()
image = x_train[num].reshape([28, 28])
plt.title('Example: %d Label: %d' % (num, label))
plt.imshow(image, cmap=plt.get_cmap('gray_r'))
plt.show()
display_digit(ran.randint(0, x_train.shape[0]))
# Import som class and train into 30 * 30 sized of SOM lattice
from som import SOM
som = SOM(30, 30, x_train.shape[1], 200)
som.train(x_train)
# Fit train data into SOM lattice
mapped = som.map_vects(x_train)
mappedarr = np.array(mapped)
x1 = mappedarr[:, 0]
y1 = mappedarr[:, 1]
index = [np.where(r == 1)[0][0] for r in y_train]
index = list(map(str, index))
## Plots: 1) Train 2) Test+Train ###
plt.figure(1, figsize=(12, 6))
plt.subplot(121)
# Plot 1 for Training only
plt.scatter(x1, y1)
# Just adding text
for i, m in enumerate(mapped):
flags = np.concatenate((flag_sig, flag_bg), axis=0) # concatenates the flag arrays into one array, each entry corresponding to the entry of the data array
### TRAINING ###
#som = SOM(dimx, dimy, 8, 400) # Train a dimx X dimy SOM with 400 iterations, 8 is the number of variables in the data array
som = SOM(dimx, dimy, 8)
som.train(data) # trains on the dataset prepared
### THIS WILL TAKE AWHILE ###
### TO STORE THE TRAINING RESULTS INTERACTIVELY IN ipython DO:
### weightages = som._weightages
### %store weightages
### THEN TO RECOVER DO:
### %store -r
### som = SOM(dimx, dimy, 8, 400)
### som._weightages = weightages
### som._trained = True
print(str(datetime.datetime.now())) # print the time to observe how long the training will take
mapped = np.array(som.map_vects(data)) # map each datapoint to its nearest neuron
# post training manipulations to the final dataset (one big numpy array)
data = np.append(data,mapped,axis=1) # append the mapped neurons to data
data = np.column_stack((data,flags)) # append the flags to the dataset
##test U-matrix
weights = som.output_weights
umatrix = get_umatrix(weights, dimx, dimy)
fig = plt.figure()
plt.imshow(umatrix, origin='lower')
plt.show(block=True)
plt.savefig('Umatrix_march29_v2.png')
##### APPLICATION #####
# get new data
### PREPARE THE SIGNAL ARRAY ###
sig2 = np.loadtxt('ksigma_mc_som_signal_feb2019.txt') # convert to np array
np.random.shuffle(sig2)
def SelfOrganizingMap(Feature_List, where):
if not os.path.exists(where):
os.makedirs(where)
som = SOM(SOM_LINES, SOM_COLS, len(Feature_List[0]), NUM_ITERATIONS)
som.train(Feature_List)
# Get output grid
image_grid = som.get_centroids()
#print 'Image grid shows weights for all the input features'
#print image_grid
# Get vector mappings
mapped = som.map_vects(Feature_List)
#print 'Mapping', mapped
# Visualization part
# Needs to be refactored to produce output for any number of clusters
avg_line = []
for i in range(NUM_CLUST):
avg_line.append(0)
avg_column = []
for i in range(NUM_CLUST):
avg_column.append(0)
for i in range(0, len(mapped), NUM_CLUST):
for j in range(NUM_CLUST):
avg_line[(i + j) % NUM_CLUST] += mapped[i + j][0]
avg_column[(i + j) % NUM_CLUST] += mapped[i + j][1]
for i in range(len(avg_line)):
avg_line[i] = avg_line[i] / (len(mapped) / NUM_CLUST)
for i in range(len(avg_column)):
avg_column[i] = avg_column[i] / (len(mapped) / NUM_CLUST)
'Average location'
print avg_line, avg_column
# Grayscale landscape
sorting_landscape = []
for i in range(NUM_CLUST):
sorting_landscape.append(np.zeros((SOM_LINES, SOM_COLS), np.float16))
for i in range(len(mapped)):
lin = mapped[i][0]
col = mapped[i][1]
sorting_landscape[i % NUM_CLUST][lin][col] += 0.05
if sorting_landscape[i % NUM_CLUST][lin][col] > 1.0:
sorting_landscape[i % NUM_CLUST][lin][col] = 1.0
for i in range(NUM_CLUST):
sorting_landscape[i][avg_line[i]][avg_column[i]] = 1.0
for i in range(NUM_CLUST):
io.imsave(where + str(i) + '_sorting_cluster.png',
sorting_landscape[i])
# Colored landscape
colored_landscape = np.zeros((SOM_LINES, SOM_COLS, 3), np.float16)
white = np.array([1.0, 1.0, 1.0])
# sorting
insertion_color = np.array([0.0, 0.0, 1.0]) # blue
bubble_color = np.array([0.0, 1.0, 0.0]) # green
heap_color = np.array([1.0, 0.0, 0.0]) # red
quick_color = np.array([1.0, 0.0, 1.0]) # magenta
random_color = np.array([1.0, 1.0, 0.0]) # yellow
# non-sorting
reverseSorted_color = np.array([0.0, 1.0, 1.0]) # cyan
intervalSwapSorted_color = np.array([1.0, 0.5, 0.0]) # orange
reverse_color = np.array([0.5, 1.0, 0.0]) # lime-green
intervalSwap_color = np.array([1.0, 0.0, 0.5]) # fuchsia
colored = [
insertion_color, bubble_color, heap_color, quick_color, random_color
]
#add_colored = [reverseSorted_color, intervalSwapSorted_color, reverse_color, intervalSwap_color]
#colored.extend(add_colored)
grad = 0.005
for i in range(len(mapped)):
lin = mapped[i][0]
col = mapped[i][1]
colored_landscape[lin][col] += grad * colored[i % NUM_CLUST]
#if colored_landscape[lin][col].any() > 1.0:
# colored_landscape[lin][col] -= 0.05*colored[i%NUM_CLUST]
for i in range(NUM_CLUST):
colored_landscape[avg_line[i]][
avg_column[i]] = 0.3 * white + 0.7 * colored[i]
io.imsave(where + 'all_sorting_clusters.png', colored_landscape)
