How do optimisers adapt to custom learning rates? - keras

I'm having difficulty in understanding how an Optimiser uses custom learning rates defined by a LearningRateScheduler. Looking at the source for SGD, the parameters seem to be affected by its own parameters only (e.g., self.lr) only:
def get_updates(self, loss, params):
grads = self.get_gradients(loss, params)
self.updates = [K.update_add(self.iterations, 1)]
lr = self.lr
if self.initial_decay > 0:
lr *= (1. / (1. + self.decay * K.cast(self.iterations,
K.dtype(self.decay))))
# momentum
shapes = [K.int_shape(p) for p in params]
moments = [K.zeros(shape) for shape in shapes]
self.weights = [self.iterations] + moments
for p, g, m in zip(params, grads, moments):
v = self.momentum * m - lr * g # velocity
self.updates.append(K.update(m, v))
if self.nesterov:
new_p = p + self.momentum * v - lr * g
else:
new_p = p + v
...
https://github.com/keras-team/keras/blob/master/keras/optimizers.py#L172
If fit_generator was called using:
def step_decay(epoch):
initial_lrate = 0.01
drop = 0.5
epochs_drop = 10.0
lrate = initial_lrate * math.pow(drop, math.floor((1+epoch)/epochs_drop))
return lrate
lr_scheduler = LearningRateScheduler(step_decay)
model.fit_generator(
...
callbacks=[lr_scheduler],
...)
is the original learning rate used for initialising the optimiser overridden somehow? It seems to be to a certain extent. I've tried the following printer:
class LearningRatePrinter(Callback):
def init(self):
super(LearningRatePrinter, self).init()
def on_epoch_begin(self, epoch, logs={}):
print('lr:', sess.run(optimizer.lr))
lr_printer = LearningRatePrinter()
and call fit_generator() with callbacks=[lr_scheduler, lr_printer,]. What I get is lr: 0.01 for each epoch. Sure, lr_printer prints the member variable lr, but how is the lr_scheduler custom learning rate value used at all? There's no sign that it's actually being used.

Related

RuntimeError: Error(s) in loading state_dict for Actor - torch.load()

I have created a custom environment in open ai gym and i am facing error while loading the weights Could some one help me to resolve the issue . I am training a TD3 network in a custom environment and i have trained successfully but while inferencing i am facing this issue
class Actor(nn.Module):
def __init__(self, state_dim, action_dim, max_action):
super(Actor, self).__init__()
self.layer_1 = nn.Linear(state_dim, 400)
self.layer_2 = nn.Linear(400, 300)
self.layer_3 = nn.Linear(300, action_dim)
self.max_action = max_action
def forward(self, x):
x = F.relu(self.layer_1(x))
x = F.relu(self.layer_2(x))
x = self.max_action * torch.tanh(self.layer_3(x))
return x
class Critic(nn.Module):
def __init__(self, state_dim, action_dim):
super(Critic, self).__init__()
# Defining the first Critic neural network
self.layer_1 = nn.Linear(state_dim + action_dim, 400)
self.layer_2 = nn.Linear(400, 300)
self.layer_3 = nn.Linear(300, 1)
# Defining the second Critic neural network
self.layer_4 = nn.Linear(state_dim + action_dim, 400)
self.layer_5 = nn.Linear(400, 300)
self.layer_6 = nn.Linear(300, 1)
def forward(self, x, u):
xu = torch.cat([x, u], 1)
# Forward-Propagation on the first Critic Neural Network
x1 = F.relu(self.layer_1(xu))
x1 = F.relu(self.layer_2(x1))
x1 = self.layer_3(x1)
# Forward-Propagation on the second Critic Neural Network
x2 = F.relu(self.layer_4(xu))
x2 = F.relu(self.layer_5(x2))
x2 = self.layer_6(x2)
return x1, x2
def Q1(self, x, u):
xu = torch.cat([x, u], 1)
x1 = F.relu(self.layer_1(xu))
x1 = F.relu(self.layer_2(x1))
x1 = self.layer_3(x1)
return x1
# Selecting the device (CPU or GPU)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Building the whole Training Process into a class
class TD3(object):
def __init__(self, state_dim, action_dim, max_action):
self.actor = Actor(state_dim, action_dim, max_action).to(device)
self.actor_target = Actor(state_dim, action_dim, max_action).to(device)
self.actor_target.load_state_dict(self.actor.state_dict())
self.actor_optimizer = torch.optim.Adam(self.actor.parameters())
self.critic = Critic(state_dim, action_dim).to(device)
self.critic_target = Critic(state_dim, action_dim).to(device)
self.critic_target.load_state_dict(self.critic.state_dict())
self.critic_optimizer = torch.optim.Adam(self.critic.parameters())
self.max_action = max_action
def select_action(self, state):
state = torch.Tensor(state.reshape(1, -1)).to(device)
return self.actor(state).cpu().data.numpy().flatten()
def train(self, replay_buffer, iterations, batch_size=100, discount=0.99, tau=0.005, policy_noise=0.2, noise_clip=0.5, policy_freq=2):
for it in range(iterations):
# Step 4: We sample a batch of transitions (s, s’, a, r) from the memory
batch_states, batch_next_states, batch_actions, batch_rewards, batch_dones = replay_buffer.sample(batch_size)
state = torch.Tensor(batch_states).to(device)
next_state = torch.Tensor(batch_next_states).to(device)
action = torch.Tensor(batch_actions).to(device)
reward = torch.Tensor(batch_rewards).to(device)
done = torch.Tensor(batch_dones).to(device)
# Step 5: From the next state s’, the Actor target plays the next action a’
next_action = self.actor_target(next_state)
# Step 6: We add Gaussian noise to this next action a’ and we clamp it in a range of values supported by the environment
noise = torch.Tensor(batch_actions).data.normal_(0, policy_noise).to(device)
noise = noise.clamp(-noise_clip, noise_clip)
next_action = (next_action + noise).clamp(-self.max_action, self.max_action)
# Step 7: The two Critic targets take each the couple (s’, a’) as input and return two Q-values Qt1(s’,a’) and Qt2(s’,a’) as outputs
target_Q1, target_Q2 = self.critic_target(next_state, next_action)
# Step 8: We keep the minimum of these two Q-values: min(Qt1, Qt2)
target_Q = torch.min(target_Q1, target_Q2)
# Step 9: We get the final target of the two Critic models, which is: Qt = r + γ * min(Qt1, Qt2), where γ is the discount factor
target_Q = reward + ((1 - done) * discount * target_Q).detach()
# Step 10: The two Critic models take each the couple (s, a) as input and return two Q-values Q1(s,a) and Q2(s,a) as outputs
current_Q1, current_Q2 = self.critic(state, action)
# Step 11: We compute the loss coming from the two Critic models: Critic Loss = MSE_Loss(Q1(s,a), Qt) + MSE_Loss(Q2(s,a), Qt)
critic_loss = F.mse_loss(current_Q1, target_Q) + F.mse_loss(current_Q2, target_Q)
# Step 12: We backpropagate this Critic loss and update the parameters of the two Critic models with a SGD optimizer
self.critic_optimizer.zero_grad()
critic_loss.backward()
self.critic_optimizer.step()
# Step 13: Once every two iterations, we update our Actor model by performing gradient ascent on the output of the first Critic model
if it % policy_freq == 0:
actor_loss = -self.critic.Q1(state, self.actor(state)).mean()
self.actor_optimizer.zero_grad()
actor_loss.backward()
self.actor_optimizer.step()
# Step 14: Still once every two iterations, we update the weights of the Actor target by polyak averaging
for param, target_param in zip(self.critic.parameters(), self.critic_target.parameters()):
target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)
# Step 15: Still once every two iterations, we update the weights of the Critic target by polyak averaging
for param, target_param in zip(self.actor.parameters(), self.actor_target.parameters()):
target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)
# Making a save method to save a trained model
def save(self, filename, directory):
torch.save(self.actor.state_dict(), '%s/%s_actor.pth' % (directory, filename))
torch.save(self.critic.state_dict(), '%s/%s_critic.pth' % (directory, filename))
# Making a load method to load a pre-trained model
def load(self, filename, directory):
self.actor.load_state_dict(torch.load('%s/%s_actor.pth' % (directory, filename)))
self.critic.load_state_dict(torch.load('%s/%s_critic.pth' % (directory, filename)))
def evaluate_policy(policy, eval_episodes=10):
avg_reward = 0.
for _ in range(eval_episodes):
obs = env.reset()
done = False
while not done:
action = policy.select_action(np.array(obs))
obs, reward, done, _ = env.step(action)
avg_reward += reward
avg_reward /= eval_episodes
print ("---------------------------------------")
print ("Average Reward over the Evaluation Step: %f" % (avg_reward))
print ("---------------------------------------")
return avg_reward
env_name = "Pygame-v0"
seed = 0
file_name = "%s_%s_%s" % ("TD3", env_name, str(seed))
print ("---------------------------------------")
print ("Settings: %s" % (file_name))
print ("---------------------------------------")
eval_episodes = 10
save_env_vid = True
env = gym.make(env_name)
max_episode_steps = env._max_episode_steps
if save_env_vid:
env = wrappers.Monitor(env, monitor_dir, force = True)
env.reset()
env.seed(seed)
torch.manual_seed(seed)
np.random.seed(seed)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
policy = TD3(state_dim, action_dim, max_action)
#policy.load(file_name, './pytorch_models/')
policy.load(file_name,"/content/gdrive/My Drive/reinforce/gym_game/pytorch_models")
_ = evaluate_policy(policy, eval_episodes=eval_episodes)
Traceback:
I am facing a runtime error while loading the state_dict for actor model .I searched google but couldnt find similar issues .
RuntimeError: Error(s) in loading state_dict for Actor:
Missing key(s) in state_dict: "layer_1.weight", "layer_1.bias", "layer_2.weight", "layer_2.bias", "layer_3.weight", "layer_3.bias".
Unexpected key(s) in state_dict: "encoder.0.weight", "encoder.0.bias", "encoder.2.weight", "encoder.2.bias", "encoder.2.running_mean", "encoder.2.running_var", "encoder.2.num_batches_tracked", "encoder.3.weight", "encoder.3.bias", "encoder.5.weight", "encoder.5.bias", "encoder.5.running_mean", "encoder.5.running_var", "encoder.5.num_batches_tracked", "encoder.6.weight", "encoder.6.bias", "encoder.8.weight", "encoder.8.bias", "encoder.8.running_mean", "encoder.8.running_var", "encoder.8.num_batches_tracked", "encoder.10.weight", "encoder.10.bias", "encoder.12.weight", "encoder.12.bias", "encoder.12.running_mean", "encoder.12.running_var", "encoder.12.num_batches_tracked", "encoder.13.weight", "encoder.13.bias", "encoder.15.weight", "encoder.15.bias", "encoder.15.running_mean", "encoder.15.running_var", "encoder.15.num_batches_tracked", "encoder.16.weight", "encoder.16.bias", "linear.0.weight", "linear.0.bias", "linear.2.weight", "linear.2.bias".
it was answered by #MicaelJungo
The weights you saved were not from the model you are using here. Make sure to load the correct checkpoint, which was created when training this particular model.

How to create my own loss function in Pytorch?

I'd like to create a model that predicts parameters of a circle (coordinates of center, radius).
Input is an array of points (of arc with noise):
def generate_circle(x0, y0, r, start_angle, phi, N, sigma):
theta = np.linspace(start_angle*np.pi/180, (start_angle + phi)*np.pi/180, num=N)
x = np.array([np.random.normal(r*np.cos(t) + x0 , sigma, 1)[0] for t in theta])
y = np.array([np.random.normal(r*np.sin(t) + y0 , sigma, 1)[0] for t in theta])
return x, y
n_x = 1000
start_angle = 0
phi = 90
N = 100
sigma = 0.005
x_full = []
for i in range(n_x):
x0 = np.random.normal(0 , 10, 1)[0]
y0 = np.random.normal(0 , 10, 1)[0]
r = np.random.normal(0 , 10, 1)[0]
x, y = generate_circle(x0, y0, r, start_angle, phi, N, sigma)
x_full.append(np.array([ [x[i], y[i]] for i in range(len(x))]))
X = torch.from_numpy(np.array(x_full))
print(X.size()) # torch.Size([1000, 100, 2])
Output: [x_c, y_c, r]
As a loss function I need to use this one:
I tried to implement something like the following:
class Net(torch.nn.Module):
def __init__(self, n_feature, n_hidden, n_output):
super(Net, self).__init__()
self.hidden = torch.nn.Linear(n_feature, n_hidden)
self.predict = torch.nn.Linear(n_hidden, n_output)
def forward(self, x):
x = F.relu(self.hidden(x))
x = self.predict(x)
return x
# It doesn't work, it's just an idea
def my_loss(point, params):
arr = ((point[:, 0] - params[:, 0])**2 + (point[:, 1] - params[:, 1])**2 - params[:, 2]**2)**2
loss = torch.sum(arr)
return loss
# For N pairs (x, y) model predicts parameters of circle
net = Net(n_feature=N*2, n_hidden=10, n_output=3)
optimizer = torch.optim.SGD(net.parameters(), lr=1e-4)
for t in range(1000):
prediction = net(X.view(n_x, N*2).float())
loss = my_loss(X, prediction)
print(f"loss: {loss}")
optimizer.zero_grad()
loss.backward()
optimizer.step()
So, the question is how to correctly implement my own loss function in terms of Pytorch in this case?
Or how to change the model's structure to get expected results?
You're trying to create a loss between the predicted outputs and the inputs instead of between the predicted outputs and the true outputs. To do this you need to save the true values of x0, y0, and r when you generate them.
n_x = 1000
start_angle = 0
phi = 90
N = 100
sigma = 0.005
x_full = []
targets = [] # <-- Here
for i in range(n_x):
x0 = np.random.normal(0 , 10, 1)[0]
y0 = np.random.normal(0 , 10, 1)[0]
r = np.random.normal(0 , 10, 1)[0]
targets.append(np.array([x0, y0, r])) # <-- Here
x, y = generate_circle(x0, y0, r, start_angle, phi, N, sigma)
x_full.append(np.array([ [x[i], y[i]] for i in range(len(x))]))
X = torch.from_numpy(np.array(x_full))
Y = torch.from_numpy(np.array(targets)) # <-- Here
print(X.size()) # torch.Size([1000, 100, 2])
print(Y.size()) # torch.Size([1000, 3])
Now, when you call my_loss you should use:
loss = my_loss(Y, prediction)
You are passing in all your data points every iteration of your for loop, I would split your data into smaller sections so that your model doesn't just learn to output the same values every time. e.g. you have generated 1000 points so pass in a random selection of 100 in each iteration using something like random.sample(...)
Your input numbers are pretty large which means your loss will be huge, so generate inputs between 0 and 1 and then if you need the value to be between 0 and 10 you can just multiply by 10.

How to resolve value error in Scipy function fmintnc?

I am trying to implement coursera assignments in python, while doing Scipy optimise for logistic regression. However, I am getting the error below.
Can any one help!
Note: cost, gradient functions are working fine.
#Sigmoid function
def sigmoid(z):
h_of_z = np.zeros([z.shape[0]])
h_of_z = np.divide(1,(1+(np.exp(-z))))
return h_of_z
def cost(x,y,theta):
m = y.shape[0]
h_of_x = sigmoid(np.matmul(x,theta))
term1 = sum(-1 * y.T # np.log(h_of_x) - (1-y.T) # np.log(1-h_of_x))
J = 1/m * term1
return J
def grad(x,y,theta):
grad = np.zeros_like(theta)
m = y.shape[0]
h_of_x = sigmoid(x#theta)
grad = (x.T # (h_of_x - y)) * (1/m)
return grad
#add intercept term for X
x = np.hstack([np.ones_like(y),X[:,0:2]])
#initialise theta
[m,n] = np.shape(x)
initial_theta = np.zeros([n,1])
#optimising theta from given theta and gradient
result = opt.fmin_tnc(func=cost, x0=initial_theta, args=(x, y))
ValueError: matmul: Input operand 1 has a mismatch in its core dimension 0, with gufunc signature (n?,k),(k,m?)->(n?,m?) (size 99 is different from 3)
I got it !
so the problem is fmin_tnc function programmed in a way we should parse the the parameter 'theta' before calling arguments x and y .
Since in my function 'cost' I have passed x and y first, it interpreted values differently so thrown ValueError .
Below are the corrected code..
def sigmoid(x):
return 1/(1+np.exp(-x))
def cost(theta,x,y):
J = (-1/m) * np.sum(np.multiply(y, np.log(sigmoid(x # theta)))
+ np.multiply((1-y), np.log(1 - sigmoid(x # theta))))
return J
def gradient(theta,x,y):
h_of_x = sigmoid(x#theta)
grad = 1 / m * (x.T # (h_of_x - y))
return grad
#initialise theta
init_theta = np.zeros([n+1,1])
#optimise theta
from scipy import optimize as op
result = op.fmin_tnc(func=cost,
x0=init_theta.flatten(),
fprime=gradient,
args=(x,y.flatten()))

Why tensorflow1.1 gets slower and slower when training? Is it memory leak or queue starvation?

I trained a ESPCN in tensorflow1.1, the costed time per patch increase nearly linearly when training. The first 100 epoch takes only 4-5 seconds, but the 70th epoch takes about half a minute. See the training result below:
I've searched the same question on Google and Stack-overflow, and tried the solutions below, but seemed no work:
1.add tf.reset_default_graph() after every sess.run();
2.add time.sleep(5) to prevent queue starvation;
I know the general idea, that is to reduce the operations in Session(). But how? Anyone have the solution?
Here's part of my code:
L3, var_w_list, var_b_list = model_train(IN, FLAGS)
cost = tf.reduce_mean(tf.reduce_sum(tf.square(OUT - L3), reduction_indices=0))
global_step = tf.Variable(0, trainable=False)
learning_rate = tf.train.exponential_decay(FLAGS.base_lr, global_step * FLAGS.batch_size, FLAGS.decay_step, 0.96, staircase=True)
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost, global_step = global_step, var_list = var_w_list + var_b_list)
# optimizer = tf.train.MomentumOptimizer(learning_rate, 0.9).minimize(cost, var_list = var_w_list + var_b_list)
cnt = 0
with tf.Session() as sess:
init_op = tf.initialize_all_variables()
sess.run(init_op)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
print('\n\n\n =========== All initialization finished, now training begins ===========\n\n\n')
t_start = time.time()
t1 = t_start
for i in range(1, FLAGS.max_Epoch + 1):
LR_batch, HR_batch = batch.__next__()
global_step += 1
[_, cost1] = sess.run([optimizer, cost], feed_dict = {IN: LR_batch, OUT: HR_batch})
# tf.reset_default_graph()
if i % 100 == 0 or i == 1:
print_step = i
print_loss = cost1 / FLAGS.batch_size
test_LR_batch, test_HR_batch = test_batch.__next__()
test_SR_batch = test_HR_batch.copy()
test_SR_batch[:,:,:,0:3] = sess.run(L3, feed_dict = {IN: test_LR_batch[:,:,:,0:3]})
# tf.reset_default_graph()
psnr_tmp = 0.0
ssim_tmp = 0.0
for k in range(test_SR_batch.shape[0]):
com1 = test_SR_batch[k, :, :, 0]
com2 = test_HR_batch[k, :, :, 0]
psnr_tmp += get_psnr(com1, com2, FLAGS.HR_size, FLAGS.HR_size)
ssim_tmp += get_ssim(com1, com2, FLAGS.HR_size, FLAGS.HR_size)
psnr[cnt] = psnr_tmp / test_SR_batch.shape[0]
ssim[cnt] = ssim_tmp / test_SR_batch.shape[0]
ep[cnt] = print_step
t2 = time.time()
print_time = t2 - t1
t1 = t2
print(("[Epoch] : {0:d} [Current cost] : {1:5.8f} \t [Validation PSNR] : {2:5.8f} \t [Duration time] : {3:10.8f} s \n").format(print_step, print_loss, psnr[cnt], print_time))
# tf.reset_default_graph()
cnt += 1
if i % 1000 == 0:
L3_test = model_test(IN_TEST, var_w_list, var_b_list, FLAGS)
output_img = single_HR.copy()
output_img[:,:,:,0:3] = sess.run(L3_test, feed_dict = {IN_TEST:single_LR[:,:,:,0:3]})
tf.reset_default_graph()
subname = FLAGS.img_save_dir + '/' + str(i) + ".jpg"
img_gen(output_img[0,:,:,:], subname)
print(('================= Saving model to {}/model.ckpt ================= \n').format(FLAGS.checkpoint_dir))
time.sleep(5)
# saver.save(sess, FLAGS.checkpoint_dir + '/model.ckpt', print_step)
t_tmp = time.time() - t_start
My configuration is: windows10 + tf1.1 + python3.5 + cuda8.0 + cudnn5.1
================================================================
Besides, I used pixel-shuffle(PS) layer instead of deconvolution in the last layer. I copied the PS code from others, which is shown below:
def _phase_shift(I, r):
bsize, a, b, c = I.get_shape().as_list()
bsize = tf.shape(I)[0] # Handling Dimension(None) type for undefined batch dim
X = tf.reshape(I, (bsize, a, b, r, r))
X = tf.transpose(X, (0, 1, 2, 4, 3)) # bsize, a, b, 1, 1
X = tf.split(X, a, 1) # a, [bsize, b, r, r]
X = tf.concat([tf.squeeze(x, axis=1) for x in X], 2) # bsize, b, a*r, r
X = tf.split(X, b, 1) # b, [bsize, a*r, r]
X = tf.concat([tf.squeeze(x, axis=1) for x in X], 2) # bsize, a*r, b*r
return tf.reshape(X, (bsize, a*r, b*r, 1))
def PS(X, r, color=False):
if color:
Xc = tf.split(X, 3, 3)
X = tf.concat([_phase_shift(x, r) for x in Xc], 3)
else:
X = _phase_shift(X, r)
return X
Which X is the 4-dimensional image tensor, r means the up-scaling factor, color determine whether the channel of images is 3(Ycbcr format) or 1(Grayscale format).
To use the layer is very simple, just like the tf.nn.relu() does:
L3_ps = PS(L3, scale, True)
Now I'm wondering whether this layer caused the slowing-down, because the program goes well when using deconvolution layer. Using deconvolution layer may be a solution, but I have to use PS layer for some reason.
I suspect this line is causing a memory leak (although without seeing the code, I can't say for certain):
L3_test = model_test(IN_TEST, var_w_list, var_b_list, FLAGS)
L3_test seems to be a tf.Tensor (because you later pass it to sess.run(), so it seems likely that model_test() is adding new nodes to the graph each time it is called (every 1000 steps), which causes more work to be done over time.
The solution is quite simple though: since model_test() does not depend on anything calculated in the training loop, you can move the call to outside the training loop, so it is only called once.

Backpropagation neural network

I need to use Backpropagation Neural Netwrok for multiclass classification purposes in my application. I have found this code and try to adapt it to my needs. It is based on the lections of Machine Learning in Coursera from Andrew Ng.
I have tested it in IRIS dataset and achieved good results (accuracy of classification around 0.96), whereas on my real data I get terrible results. I assume there is some implementation error, because the data is very simple. But I cannot figure out what exactly is the problem.
What are the parameters that it make sense to adjust?
I tried with:
number of units in hidden layer
generalization parameter (lambda)
number of iterations for minimization function
Built-in minimization function used in this code is pretty much confusing me. It is used just once, as #goncalopp has mentioned in comment. Shouldn't it iteratively update the weights? How it can be implemented?
Here is my training data (target class is in the last column):
65535, 3670, 65535, 3885, -0.73, 1
65535, 3962, 65535, 3556, -0.72, 1
65535, 3573, 65535, 3529, -0.61, 1
3758, 3123, 4117, 3173, -0.21, 0
3906, 3119, 4288, 3135, -0.28, 0
3750, 3073, 4080, 3212, -0.26, 0
65535, 3458, 65535, 3330, -0.85, 2
65535, 3315, 65535, 3306, -0.87, 2
65535, 3950, 65535, 3613, -0.84, 2
65535, 32576, 65535, 19613, -0.35, 3
65535, 16657, 65535, 16618, -0.37, 3
65535, 16657, 65535, 16618, -0.32, 3
The dependencies are so obvious, I think it should be so easy to classify it...
But results are terrible. I get accuracy of 0.6 to 0.8. This is absolutely inappropriate for my application. Can someone please point out possible improvements I could make in order to achieve better results.
Here is the code:
import numpy as np
from scipy import optimize
from sklearn import cross_validation
from sklearn.metrics import accuracy_score
import math
class NN_1HL(object):
def __init__(self, reg_lambda=0, epsilon_init=0.12, hidden_layer_size=25, opti_method='TNC', maxiter=500):
self.reg_lambda = reg_lambda
self.epsilon_init = epsilon_init
self.hidden_layer_size = hidden_layer_size
self.activation_func = self.sigmoid
self.activation_func_prime = self.sigmoid_prime
self.method = opti_method
self.maxiter = maxiter
def sigmoid(self, z):
return 1 / (1 + np.exp(-z))
def sigmoid_prime(self, z):
sig = self.sigmoid(z)
return sig * (1 - sig)
def sumsqr(self, a):
return np.sum(a ** 2)
def rand_init(self, l_in, l_out):
self.epsilon_init = (math.sqrt(6))/(math.sqrt(l_in + l_out))
return np.random.rand(l_out, l_in + 1) * 2 * self.epsilon_init - self.epsilon_init
def pack_thetas(self, t1, t2):
return np.concatenate((t1.reshape(-1), t2.reshape(-1)))
def unpack_thetas(self, thetas, input_layer_size, hidden_layer_size, num_labels):
t1_start = 0
t1_end = hidden_layer_size * (input_layer_size + 1)
t1 = thetas[t1_start:t1_end].reshape((hidden_layer_size, input_layer_size + 1))
t2 = thetas[t1_end:].reshape((num_labels, hidden_layer_size + 1))
return t1, t2
def _forward(self, X, t1, t2):
m = X.shape[0]
ones = None
if len(X.shape) == 1:
ones = np.array(1).reshape(1,)
else:
ones = np.ones(m).reshape(m,1)
# Input layer
a1 = np.hstack((ones, X))
# Hidden Layer
z2 = np.dot(t1, a1.T)
a2 = self.activation_func(z2)
a2 = np.hstack((ones, a2.T))
# Output layer
z3 = np.dot(t2, a2.T)
a3 = self.activation_func(z3)
return a1, z2, a2, z3, a3
def function(self, thetas, input_layer_size, hidden_layer_size, num_labels, X, y, reg_lambda):
t1, t2 = self.unpack_thetas(thetas, input_layer_size, hidden_layer_size, num_labels)
m = X.shape[0]
Y = np.eye(num_labels)[y]
_, _, _, _, h = self._forward(X, t1, t2)
costPositive = -Y * np.log(h).T
costNegative = (1 - Y) * np.log(1 - h).T
cost = costPositive - costNegative
J = np.sum(cost) / m
if reg_lambda != 0:
t1f = t1[:, 1:]
t2f = t2[:, 1:]
reg = (self.reg_lambda / (2 * m)) * (self.sumsqr(t1f) + self.sumsqr(t2f))
J = J + reg
return J
def function_prime(self, thetas, input_layer_size, hidden_layer_size, num_labels, X, y, reg_lambda):
t1, t2 = self.unpack_thetas(thetas, input_layer_size, hidden_layer_size, num_labels)
m = X.shape[0]
t1f = t1[:, 1:]
t2f = t2[:, 1:]
Y = np.eye(num_labels)[y]
Delta1, Delta2 = 0, 0
for i, row in enumerate(X):
a1, z2, a2, z3, a3 = self._forward(row, t1, t2)
# Backprop
d3 = a3 - Y[i, :].T
d2 = np.dot(t2f.T, d3) * self.activation_func_prime(z2)
Delta2 += np.dot(d3[np.newaxis].T, a2[np.newaxis])
Delta1 += np.dot(d2[np.newaxis].T, a1[np.newaxis])
Theta1_grad = (1 / m) * Delta1
Theta2_grad = (1 / m) * Delta2
if reg_lambda != 0:
Theta1_grad[:, 1:] = Theta1_grad[:, 1:] + (reg_lambda / m) * t1f
Theta2_grad[:, 1:] = Theta2_grad[:, 1:] + (reg_lambda / m) * t2f
return self.pack_thetas(Theta1_grad, Theta2_grad)
def fit(self, X, y):
num_features = X.shape[0]
input_layer_size = X.shape[1]
num_labels = len(set(y))
theta1_0 = self.rand_init(input_layer_size, self.hidden_layer_size)
theta2_0 = self.rand_init(self.hidden_layer_size, num_labels)
thetas0 = self.pack_thetas(theta1_0, theta2_0)
options = {'maxiter': self.maxiter}
_res = optimize.minimize(self.function, thetas0, jac=self.function_prime, method=self.method,
args=(input_layer_size, self.hidden_layer_size, num_labels, X, y, 0), options=options)
self.t1, self.t2 = self.unpack_thetas(_res.x, input_layer_size, self.hidden_layer_size, num_labels)
np.savetxt("weights_t1.txt", self.t1, newline="\n")
np.savetxt("weights_t2.txt", self.t2, newline="\n")
def predict(self, X):
return self.predict_proba(X).argmax(0)
def predict_proba(self, X):
_, _, _, _, h = self._forward(X, self.t1, self.t2)
return h
##################
# IR data #
##################
values = np.loadtxt('infrared_data.txt', delimiter=', ', usecols=[0,1,2,3,4])
targets = np.loadtxt('infrared_data.txt', delimiter=', ', dtype=(int), usecols=[5])
X_train, X_test, y_train, y_test = cross_validation.train_test_split(values, targets, test_size=0.4)
nn = NN_1HL()
nn.fit(values, targets)
print("Accuracy of classification: "+str(accuracy_score(y_test, nn.predict(X_test))))
The most obvious problem is that your training dataset is very small.
Since you're using scipy.optimize.minimize instead of the usual iterative gradient descent, I think it's also likely you're overfitting your model to your training data. Possibly a iterative algorithm works better, here. Don't forget to carefully monitor the validation error.
If you try backpropagation with gradient descent, notice that, depending on the parameters used on backpropagation, neural networks take a while to converge
You can try to feed the network the same training data multiple times or tweak the learning rate but ideally you should use more diverse data.
Correctly normalizing the data solved the problem. I used preprocessing module from sklearn. Here is example:
from sklearn import preprocessing
import numpy as np
X_train = np.array([[ 1., -1., 2.],
[ 2., 0., 0.],
[ 0., 1., -1.]])
min_max_scaler = preprocessing.MinMaxScaler()
X_train_minmax = min_max_scaler.fit_transform(X_train)
print(X_train_minmax)
X_test = np.array([[ -3., -1., 4.]])
X_test_minmax = min_max_scaler.transform(X_test)
print(X_test_minmax)
And the output is:
[[ 0.5 0. 1. ]
[ 1. 0.5 0.3333]
[ 0. 1. 0. ]]
[[-1.5 0. 1.6667]]

Resources