def sim2(systems):
keynames, keys, data = prep_data('M', systems)
for i in range(10):
som = SOM(30, 30, 12, .25, systems, 'M')
som.train(100, data,
'../LogFiles/Sim2/' + systems + '/log' + str(i) + '.json')
output_results(som, keynames, keys)
def main():
'''CONTROL PANEL'''
MODE = 'K' # 'K' to train on musical keys, 'M' to train on music datasets
SYSTEMS = 'H' # Include 'W' for Western system, 'C' for Chinese system, 'H' for Hindustani system
MODE2 = 'M'
SYSTEMS2 = 'HC'
width = 30
height = 30
iterations = 100
learning_rate = .25
pitch_count = 12
'''================'''
# Generate model
#print "Generating map..."
som = SOM(width, height, pitch_count, learning_rate, SYSTEMS, MODE)
genTest(som, iterations, MODE, SYSTEMS)
def sim3(mode1, mode2, sys1, sys2):
for i in range(10):
f = open('../LogFiles/Sim1/' + sys1 + '/log' + str(i) + '.json', 'r')
json = jlb(f)
nodes = scipy.array(json['states'][-1])
som = SOM(30, 30, 12, .25, sys2, mode2, nodes)
# Fix radius and learning rates
som.learning_rate /= 100.
som.radius = 1.
som.fixedR = True
som.fixedLR = True
# Update dataset to include new system(s)
keynames, keys, data = prep_data(mode2, sys2)
som.train(100, data,
'../LogFiles/Sim3a/' + sys2 + '/log' + str(i) + '.json')
output_results(som, keynames, keys)
def score_all():
for sys in 'WCH':
for file_name in os.listdir('../LogFiles/Sim1/' + sys):
with open('../LogFiles/Sim1/' + sys + '/' + file_name, 'r+') as f:
data = jlb(f)
states = data['states']
som = SOM(30, 30, 12, .25, sys, 'M')
for state in states:
state = np.array(state)
som.set_nodes(state)
som.score()
scores = som.log['scores']
data = {'states':states, 'scores':scores}
f.seek(0)
json.dump(data, f)
f.truncate()
# Toyset
feat_nr = 3
data = np.random.randint(0, 255, (100,3))
np.save("./data/toydata.npy", data)
data_path = "./data/toydata.npy"
##################
# set parameters #
##################
trial = "toytest"
log_path = "./data/"
num_steps = 100
x = 10
y = 10
gpu=None
#get the model
som = SOM(trial, x, y, num_steps, data_path, log_path, gpu=gpu, norm=True,
learning_rate=0.1)
#train the model
som.train()
# get the trained map
net = som.get_weights()
plot_colour(net)
y = dp.convertTextTarget(y, alias)
#dump_result(output_path + 'accidents.csv', np.array(alias), ['accident'])
print('Accident types: ', alias)
# Standardize data
X = ds.standardize(X)
print('======== Training ============')
# ============================================ #
# Train SOM
m = 20
n = 20
n_iter = 15000
som = SOM(m,
n,
len(col_name) - 1,
n_iter,
save=wd + 'checkpoint/',
restore=wd + 'checkpoint/')
som.train(X, checkpoint_len=2)
# ============================================ #
# Get cluster grid formed by SOM
cluster_grid = som.get_centroids()
cluster_grid = np.array(cluster_grid)
cluster_grid = np.reshape(cluster_grid, (m * n, len(col_name) - 1))
cluster_grid = ds.inverse_transform(cluster_grid)
# Dump data to plot on R
dump_result(output_path + 'cluster_grid.csv', cluster_grid, col_name[:-1])
cluster_grid = np.reshape(cluster_grid, (m, n, len(col_name) - 1))
##################
trial = "100x100"
data_path = "./data/instances.npy"
#log_path = "~/cache/tensorboard-logdir/jan"
log_path = "./data/"
num_steps = 100
x = 50
y = 50
gpu = None
#get the model
som = SOM(trial,
x,
y,
num_steps,
data_path,
log_path,
gpu=gpu,
norm=True,
learning_rate=0.1)
#train the model
som.train()
# get the trained map
#net = som.get_weights()
### save/load weights
som.save_weights()
#som.load_weights("./data/test.npy")