lets say I have model called UNet
output = UNet(input)
that output is a vector of grayscale images shape: (batch_size,1,128,128)
What I want to do is to normalize each image to be in range [0,1].
I did it like this:
for i in range(batch_size):
output[i,:,:,:] = output[i,:,:,:]/torch.amax(output,dim=(1,2,3))[i]
now every image in the output is normalized, but when I'm training such model, pytorch claim it cannot calculate the gradients in this procedure, and I understand why.
my question is what is the right way to normalize image without killing the backpropogation flow?
something like
output = UNet(input)
output = output.normalize
output2 = some_model(output)
loss = ..
loss.backward()
optimize.step()
my only option right now is adding a sigmoid activation at the end of the UNet but i dont think its a good idea..
update - code (gen2,disc = unet,discriminator models. est_bias is some output):
update 2x code:
with torch.no_grad():
est_bias_for_disc = gen2(input_img)
est_bias_for_disc /= est_bias_for_disc.amax(dim=(1,2,3), keepdim=True)
disc_fake_hat = disc(est_bias_for_disc.detach())
disc_fake_loss = BCE(disc_fake_hat, torch.zeros_like(disc_fake_hat))
disc_real_hat = disc(bias_ref)
disc_real_loss = BCE(disc_real_hat, torch.ones_like(disc_real_hat))
disc_loss = (disc_fake_loss + disc_real_loss) / 2
if epoch<=epochs_till_gen2_stop:
disc_loss.backward(retain_graph=True) # Update gradients
opt_disc.step() # Update optimizer
then theres seperate training:
opt_gen2.zero_grad()
est_bias = gen2(input_img)
est_bias /= est_bias.amax(dim=(1,2,3), keepdim=True)
disc_fake = disc(est_bias)
ADV_loss = BCE(disc_fake, torch.ones_like(disc_fake))
gen2_loss = ADV_loss
gen2_loss.backward()
opt_gen2.step()
You can use the normalize function:
>>> import torch
>>> import torch.nn.functional as F
>>> x = torch.tensor([[3.,4.],[5.,6.],[7.,8.]])
>>> x = F.normalize(x, dim = 0)
>>> print(x)
tensor([[0.3293, 0.3714],
[0.5488, 0.5571],
[0.7683, 0.7428]])
This will give a differentiable tensor as long as out is not used.
You are overwriting the tensor's value because of the indexing on the batch dimension. Instead, you can perform the operation in vectorized form:
output = output / output.amax(dim=(1,2,3), keepdim=True)
The keepdim=True argument keeps the shape of torch.Tensor.amax's output equal to that of its inputs allowing you to perform an in-place operation with it.
Related
I just find that in the code here:
https://github.com/NUS-Tim/Pytorch-WGAN/tree/master/models
The "generator" loss, G, between WGAN and WGAN-GP is different, for WGAN:
g_loss = self.D(fake_images)
g_loss = g_loss.mean().mean(0).view(1)
g_loss.backward(one) # !!!
g_cost = -g_loss
But for WGAN-GP:
g_loss = self.D(fake_images)
g_loss = g_loss.mean()
g_loss.backward(mone) # !!!
g_cost = -g_loss
Why one is one=1 and another is mone=-1?
You might have misread the source code, the first sample you gave is not averaging the resut of D to compute its loss but instead uses the binary cross-entropy.
To be more precise:
The first method ("GAN") uses the BCE loss to compute the loss terms for D and G. The standard GAN optimization objective for D is to minimize E_x[log(D(x))] + E_z[log(1-D(G(z)))]. Source code:
outputs = self.D(images)
d_loss_real = self.loss(outputs.flatten(), real_labels) # <- bce loss
real_score = outputs
# Compute BCELoss using fake images
fake_images = self.G(z)
outputs = self.D(fake_images)
d_loss_fake = self.loss(outputs.flatten(), fake_labels) # <- bce loss
fake_score = outputs
# Optimizie discriminator
d_loss = d_loss_real + d_loss_fake
self.D.zero_grad()
d_loss.backward()
self.d_optimizer.step()
For d_loss_real you optimize towards 1s (output is considered real), while d_loss_fake optimizes towards 0s (output is considered fake).
While the second ("WCGAN") uses the Wasserstein loss (ref) whereby we maximise for D the loss: E_x[D(x)] - E_z[D(G(z))]. Source code:
# Train discriminator
# WGAN - Training discriminator more iterations than generator
# Train with real images
d_loss_real = self.D(images)
d_loss_real = d_loss_real.mean()
d_loss_real.backward(mone)
# Train with fake images
z = self.get_torch_variable(torch.randn(self.batch_size, 100, 1, 1))
fake_images = self.G(z)
d_loss_fake = self.D(fake_images)
d_loss_fake = d_loss_fake.mean()
d_loss_fake.backward(one)
# [...]
Wasserstein_D = d_loss_real - d_loss_fake
By doing d_loss_real.backward(mone) you backpropage with a gradient of opposite sign, i.e. its's a gradient ascend, and you end up maximizing d_loss_real.
In order to Update D network:
lossD = Expectation of D(fake data) - Expectation of D(real data) + gradient penalty
lossD ↓，D(real data) ↑
so you need to add minus one to the gradient process
I am conducting some experiments on multi-label classification via deep learning models.
But I face a problem with the dataset.
I use Keras,TensorFlow 2.0, numpy,pandas.
I have a dataset in the form:
Dataset in the form that I have it
To apply multi-label classification(6 labels) I need my dataset to be in this form:
Dataset in the form that I need it
How is it possible to achieve this? Are there any functions making this transformation easier?
Try:
comments_df[['abusive','hateful','offensive','disrespectful','fearful','normal']] = comments_df['sentiment'].str.split('_', -1, expand=True)
This gives me an error:
ValueError: Columns must be same length as key
Regarding the DL model I will use, it's bi-LSTM, but it doesn't have anything to do with the question per-se.
Try this:
df = pd.get_dummies(data = df, columns = ['sentiment'])
I found this to work (not optimal solution):
"""
Creating a column for each of the target labels with sentiment's column data.
"""
def split_sentiment_outputs(output_label, sentiment_col="sentiment"):
comments_df[output_label] = comments_df[sentiment_col].str.split('_')
"""
Transform column's data to categorical.
"""
def transform_data_for_multilabel(output_label):
row = comments_df[output_label]
for index, row in row.items():
# print("Index:", index)
# print("length:", len(row))
# print("content:", row)
# print("--------------")
z = 0
while z < len(row):
if row[z] == output_label:
comments_df.at[index, output_label] = 1
break
else:
comments_df.at[index, output_label] = 0
z = z + 1
# Applying Data Transformation
output_labels = ["abusive", "hateful", "offensive", "disrespectful", "fearful", "normal"]
for i in range(MAX_OUT):
split_sentiment_outputs(output_labels[i])
for i in range(MAX_OUT):
transform_data_for_multilabel(output_labels[i])
I am new to tensorflow (and my first question in StackOverflow)
As a learning tool, I am trying to do something simple. (4 days later I am still confused)
I have one CSV file with 36 columns (3500 records) with 0s and 1s.
I am envisioning this file as a flattened 6x6 matrix.
I have another CSV file with 1 columnn of ground truth 0 or 1 (3500 records) which indicates if at least 4 of the 6 of elements in the 6x6 matrix's diagonal are 1's.
I am not sure I have processed the CSV files correctly.
I am confused as to how I create the features dictionary and Labels and how that fits into the DNNClassifier
I am using TensorFlow 1.6, Python 3.6
Below is the small amount of code I have so far.
import tensorflow as tf
import os
def x_map(line):
rDefaults = [[] for cl in range(36)]
x_row = tf.decode_csv(line, record_defaults=rDefaults)
return x_row
def y_map(line):
line = tf.string_to_number(line, out_type=tf.int32)
y_row = tf.one_hot(line, depth=2)
return y_row
x_path_file = os.path.join('D:', 'Diag', '6x6_train.csv')
y_path_file = os.path.join('D:', 'Diag', 'HasDiag_train.csv')
filenames = [x_path_file]
x_dataset = tf.data.TextLineDataset(filenames)
x_dataset = x_dataset.map(x_map)
x_dataset = x_dataset.batch(1)
x_iter = x_dataset.make_one_shot_iterator()
x_next_el = x_iter.get_next()
filenames = [y_path_file]
y_dataset = tf.data.TextLineDataset(filenames)
y_dataset = y_dataset.map(y_map)
y_dataset = y_dataset.batch(1)
y_iter = y_dataset.make_one_shot_iterator()
y_next_el = y_iter.get_next()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
x_el = (sess.run(x_next_el))
y_el = (sess.run(y_next_el))
The output for x_el is:
(array([1.], dtype=float32), array([1.], dtype=float32), array([1.], dtype=float32), array([1.], dtype=float32), array([1.], dtype=float32), array([0.] ... it goes on...
The output for y_el is:
[[1. 0.]]
You're pretty much there for a minimal working model. The main issue I see is that tf.decode_csv returns a tuple of tensors, where as I expect you want a single tensor with all values. Easy fix:
x_row = tf.stack(tf.decode_csv(line, record_defaults=rDefaults))
That should work... but it fails to take advantage of many of the awesome things the tf.data.Dataset API has to offer, like shuffling, parallel threading etc. For example, if you shuffle each dataset, those shuffling operations won't be consistent. This is because you've created two separate datasets and manipulated them independently. If you create them independently, zip them together then manipulate, those manipulations will be consistent.
Try something along these lines:
def get_inputs(
count=None, shuffle=True, buffer_size=1000, batch_size=32,
num_parallel_calls=8, x_paths=[x_path_file], y_paths=[y_path_file]):
"""
Get x, y inputs.
Args:
count: number of epochs. None indicates infinite epochs.
shuffle: whether or not to shuffle the dataset
buffer_size: used in shuffle
batch_size: size of batch. See outputs below
num_parallel_calls: used in map. Note if > 1, intra-batch ordering
will be shuffled
x_paths: list of paths to x-value files.
y_paths: list of paths to y-value files.
Returns:
x: (batch_size, 6, 6) tensor
y: (batch_size, 2) tensor of 1-hot labels
"""
def x_map(line):
rDefaults = [[] for cl in range(n_dims**2)]
x_row = tf.stack(tf.decode_csv(line, record_defaults=rDefaults))
return x_row
def y_map(line):
line = tf.string_to_number(line, out_type=tf.int32)
y_row = tf.one_hot(line, depth=2)
return y_row
def xy_map(x, y):
return x_map(x), y_map(y)
x_ds = tf.data.TextLineDataset(x_paths)
y_ds = tf.data.TextLineDataset(y_paths)
combined = tf.data.Dataset.zip((x_ds, y_ds))
combined = combined.repeat(count=count)
if shuffle:
combined = combined.shuffle(buffer_size)
combined = combined.map(xy_map, num_parallel_calls=num_parallel_calls)
combined = combined.batch(batch_size)
x, y = combined.make_one_shot_iterator().get_next()
return x, y
To experiment/debug,
x, y = get_inputs()
with tf.Session() as sess:
xv, yv = sess.run((x, y))
print(xv.shape, yv.shape)
For use in an estimator, pass the function itself.
estimator.train(get_inputs, max_steps=10000)
def get_eval_inputs():
return get_inputs(
count=1, shuffle=False
x_paths=[x_eval_paths],
y_paths=[y_eval_paths])
estimator.eval(get_eval_inputs)
I am trying to implement q-learning with an action-value approximation-function. I am using openai-gym and the "MountainCar-v0" enviroment to test my algorithm out. My problem is, it does not converge or find the goal at all.
Basically the approximator works like the following, you feed in the 2 features: position and velocity and one of the 3 actions in a one-hot encoding: 0 -> [1,0,0], 1 -> [0,1,0] and 2 -> [0,0,1]. The output is the action-value approximation Q_approx(s,a), for one specific action.
I know that usually, the input is the state (2 features) and the output layer contains 1 output for each action. The big difference that I see is that I have run the feed forward pass 3 times (one for each action) and take the max, while in the standard implementation you run it once and take the max over the output.
Maybe my implementation is just completely wrong and I am thinking wrong. Gonna paste the code here, it is a mess but I am just experimenting a bit:
import gym
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, Activation
env = gym.make('MountainCar-v0')
# The mean reward over 20 episodes
mean_rewards = np.zeros(20)
# Feature numpy holder
features = np.zeros(5)
# Q_a value holder
qa_vals = np.zeros(3)
one_hot = {
0 : np.asarray([1,0,0]),
1 : np.asarray([0,1,0]),
2 : np.asarray([0,0,1])
}
model = Sequential()
model.add(Dense(20, activation="relu",input_dim=(5)))
model.add(Dense(10,activation="relu"))
model.add(Dense(1))
model.compile(optimizer='rmsprop',
loss='mse',
metrics=['accuracy'])
epsilon_greedy = 0.1
discount = 0.9
batch_size = 16
# Experience replay containing features and target
experience = np.ones((10*300,5+1))
# Ring buffer
def add_exp(features,target,index):
if index % experience.shape[0] == 0:
index = 0
global filled_once
filled_once = True
experience[index,0:5] = features
experience[index,5] = target
index += 1
return index
for e in range(0,100000):
obs = env.reset()
old_obs = None
new_obs = obs
rewards = 0
loss = 0
for i in range(0,300):
if old_obs is not None:
# Find q_a max for s_(t+1)
features[0:2] = new_obs
for i,pa in enumerate([0,1,2]):
features[2:5] = one_hot[pa]
qa_vals[i] = model.predict(features.reshape(-1,5))
rewards += reward
target = reward + discount*np.max(qa_vals)
features[0:2] = old_obs
features[2:5] = one_hot[a]
fill_index = add_exp(features,target,fill_index)
# Find new action
if np.random.random() < epsilon_greedy:
a = env.action_space.sample()
else:
a = np.argmax(qa_vals)
else:
a = env.action_space.sample()
obs, reward, done, info = env.step(a)
old_obs = new_obs
new_obs = obs
if done:
break
if filled_once:
samples_ids = np.random.choice(experience.shape[0],batch_size)
loss += model.train_on_batch(experience[samples_ids,0:5],experience[samples_ids,5].reshape(-1))[0]
mean_rewards[e%20] = rewards
print("e = {} and loss = {}".format(e,loss))
if e % 50 == 0:
print("e = {} and mean = {}".format(e,mean_rewards.mean()))
Thanks in advance!
There shouldn't be much difference between the actions as inputs to your network or as different outputs of your network. It does make a huge difference if your states are images for example. because Conv nets work very well with images and there would be no obvious way of integrating the actions to the input.
Have you tried the cartpole balancing environment? It is better to test if your model is working correctly.
Mountain climb is pretty hard. It has no reward until you reach the top, which often doesn't happen at all. The model will only start learning something useful once you get to the top once. If you are never getting to the top you should probably increase your time doing exploration. in other words take more random actions, a lot more...
I am trying to understand how FiPy works by working an example, in particular I would like to solve the following simple convection equation with periodic boundary:
$$\partial_t u + \partial_x u = 0$$
If initial data is given by $u(x, 0) = F(x)$, then the analytical solution is $u(x, t) = F(x - t)$. I do get a solution, but it is not correct.
What am I missing? Is there a better resource for understanding FiPy than the documentation? It is very sparse...
Here is my attempt
from fipy import *
import numpy as np
# Generate mesh
nx = 20
dx = 2*np.pi/nx
mesh = PeriodicGrid1D(nx=nx, dx=dx)
# Generate solution object with initial discontinuity
phi = CellVariable(name="solution variable", mesh=mesh)
phiAnalytical = CellVariable(name="analytical value", mesh=mesh)
phi.setValue(1.)
phi.setValue(0., where=x > 1.)
# Define the pde
D = [[-1.]]
eq = TransientTerm() == ConvectionTerm(coeff=D)
# Set discretization so analytical solution is exactly one cell translation
dt = 0.01*dx
steps = 2*int(dx/dt)
# Set the analytical value at the end of simulation
phiAnalytical.setValue(np.roll(phi.value, 1))
for step in range(steps):
eq.solve(var=phi, dt=dt)
print(phi.allclose(phiAnalytical, atol=1e-1))
As addressed on the FiPy mailing list, FiPy is not great at handling convection only PDEs (absent diffusion, pure hyperbolic) as it's missing higher order convection schemes. It is better to use CLAWPACK for this class of problem.
FiPy does have one second order scheme that might help with this problem, the VanLeerConvectionTerm, see an example.
If the VanLeerConvectionTerm is used in the above problem, it does do a better job of preserving the shock.
import numpy as np
import fipy
# Generate mesh
nx = 20
dx = 2*np.pi/nx
mesh = fipy.PeriodicGrid1D(nx=nx, dx=dx)
# Generate solution object with initial discontinuity
phi = fipy.CellVariable(name="solution variable", mesh=mesh)
phiAnalytical = fipy.CellVariable(name="analytical value", mesh=mesh)
phi.setValue(1.)
phi.setValue(0., where=mesh.x > 1.)
# Define the pde
D = [[-1.]]
eq = fipy.TransientTerm() == fipy.VanLeerConvectionTerm(coeff=D)
# Set discretization so analytical solution is exactly one cell translation
dt = 0.01*dx
steps = 2*int(dx/dt)
# Set the analytical value at the end of simulation
phiAnalytical.setValue(np.roll(phi.value, 1))
viewer = fipy.Viewer(phi)
for step in range(steps):
eq.solve(var=phi, dt=dt)
viewer.plot()
raw_input('stopped')
print(phi.allclose(phiAnalytical, atol=1e-1))