Develop Reference

Problems with Python tensorflow

I am a noob in programming who tried to study machine learning. I used tensorflow for Python. Here's the code, written (but not 100% copied) with official tensorflow guide (here's it https://www.tensorflow.org/guide/basics). I can't see the final graph with the results after training. I've tried two methods of training and both share the same problem. Could anyone help me?
import matplotlib as mp
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as pl
mp.rcParams["figure.figsize"] = [20, 10]
precision = 500
x = tf.linspace(-10.0, 10.0, precision)
def y(x): return 4 * np.sin(x - 1) + 3
newY = y(x) + tf.random.normal(shape=[precision])
class Model(tf.keras.Model):
def __init__(self, units):
super().__init__()
self.dense1 = tf.keras.layers.Dense(units = units, activation = tf.nn.relu, kernel_initializer=tf.random.normal, bias_initializer=tf.random.normal)
self.dense2 = tf.keras.layers.Dense(1)
def __call__(self, x, training = True):
x = x[:, tf.newaxis]
x = self.dense1(x)
x = self.dense2(x)
return tf.squeeze(x, axis=1)
model = Model(164)
pl.plot(x, y(x), label = "origin")
pl.plot(x, newY, ".", label = "corrupted")
pl.plot(x, model(x), label = "before training")
""" The first method
vars = model.variables
optimizer = tf.optimizers.SGD(learning_rate = 0.01)
for i in range(1000):
with tf.GradientTape() as tape:
prediction = model(x)
error = (newY-prediction)**2
mean_error = tf.reduce_mean(error)
gradient = tape.gradient(mean_error, vars)
optimizer.apply_gradients(zip(gradient, vars))
"""
model.compile(loss = tf.keras.losses.MSE, optimizer = tf.optimizers.SGD(learning_rate = 0.01))
model.fit(x, newY, epochs=100,batch_size=32,verbose=0)
pl.plot(x, model(x), label = "after training")
pl.legend()
pl.show()

I copied your code and investigated it. Your model returns NaN loss during training, I removed kernel and bias initializers and it works. For now I don't know what's wrong with your initialization. It seems that some weights got initialized with NaNs which then made the predictions become NaNs, hence you couldn't plot them.
Update: use the initializers module (like tensorflow.initializers or tensorflow.keras.initializers, not tensorflow.random). For example, use kernel_initializer=tf.initializers.random_normal instead of what you have.

As I can see, your third graph and your fourth graph are the same. They are
pl.plot(x, model(x), label = "before training") and pl.plot(x, model(x), label = "after training") You can figure out that the x-axis and y-axis data of two graphs are the same.
Hope my answer is helpful to you!

Accuracy very bad in tensorflow logistic regression

I am trying to write a program that predicts if one has malignant tumor or benign tumor
Dataset is from:https://archive.ics.uci.edu/ml/datasets/Breast+Cancer+Wisconsin+%28Prognostic%29
This is my code and my accuracy is at about 65% which is no better than a coin flip. Any help would be appreciated
import tensorflow as tf
import pandas as pd
import numpy as np
df = pd.read_csv(r'D:\wholedesktop\logisticReal.txt')
df.drop(['id'], axis=1, inplace=True)
x_data = np.array(df.drop(['class'], axis=1))
x_data = x_data.astype(np.float64)
y = df['class']
y.replace(2, 0, inplace=True)
y.replace(4, 1, inplace=True)
y_data = np.array(y)
# y shape = 681,1
# x shape = 681,9
x = tf.placeholder(name='x', dtype=np.float32)
y = tf.placeholder(name='y', dtype=np.float32)
w = tf.Variable(dtype=np.float32, initial_value=np.random.random((9, 1)))
b = tf.Variable(dtype=np.float32, initial_value=np.random.random((1, 1)))
y_ = (tf.add(tf.matmul(x, w), b))
error = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=y_, labels=y))
goal = tf.train.GradientDescentOptimizer(0.05).minimize(error)
prediction = tf.round(tf.sigmoid(y_))
correct = tf.cast(tf.equal(prediction, y), dtype=np.float64)
accuracy = tf.reduce_mean(correct)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(2000):
sess.run(goal, feed_dict={x: x_data, y: y_data})
print(i, sess.run(accuracy, feed_dict={x: x_data, y: y_data}))
weight = sess.run(w)
bias = sess.run(b)
print(weight)
print(bias)

Your neural network only has a single layer, so the best it can do is fit a straight line to your data that separates the different classes. This is vastly insufficient for a general (high-dimensional) data set. The power of (deep) neural networks lies in the connectivity between many layers of neurons. In your example, you could add more layers manually by passing the output of matmul to a new matmul with different weights and biases, or you could use the contrib.layers collection to make it more concise:
x = tf.placeholder(name='x', dtype=np.float32)
fc1 = tf.contrib.layers.fully_connected(inputs=x, num_outputs=16, activation_fn=tf.nn.relu)
fc2 = tf.contrib.layers.fully_connected(inputs=fc1, num_outputs=32, activation_fn=tf.nn.relu)
fc3 = tf.contrib.layers.fully_connected(inputs=fc2, num_outputs=64, activation_fn=tf.nn.relu)
The trick is to pass the output from one layer as input to the next layer. As you add more and more layers, your accuracy will go up (probably because of over-fitting, use dropout to remedy that).

Difference between GradientDescentOptimizer and AdamOptimizer in tensorflow?

When using GradientDescentOptimizer instead of Adam Optimizer the model doesn't seem to converge. On the otherhand, AdamOptimizer seems to work fine. Is the something wrong with the GradientDescentOptimizer from tensorflow?
import matplotlib.pyplot as plt
import tensorflow as tf
import numpy as np
def randomSample(size=100):
"""
y = 2 * x -3
"""
x = np.random.randint(500, size=size)
y = x * 2 - 3 - np.random.randint(-20, 20, size=size)
return x, y
def plotAll(_x, _y, w, b):
fig = plt.figure()
ax = fig.add_subplot(111)
ax.scatter(_x, _y)
x = np.random.randint(500, size=20)
y = w * x + b
ax.plot(x, y,'r')
plt.show()
def lr(_x, _y):
w = tf.Variable(2, dtype=tf.float32)
b = tf.Variable(3, dtype=tf.float32)
x = tf.placeholder(tf.float32)
y = tf.placeholder(tf.float32)
linear_model = w * x + b
loss = tf.reduce_sum(tf.square(linear_model - y))
optimizer = tf.train.AdamOptimizer(0.0003) #GradientDescentOptimizer
train = optimizer.minimize(loss)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
for i in range(10000):
sess.run(train, {x : _x, y: _y})
cw, cb, closs = sess.run([w, b, loss], {x:_x, y:_y})
print(closs)
print(cw,cb)
return cw, cb
x,y = randomSample()
w,b = lr(x,y)
plotAll(x,y, w, b)

I had a similar problem once and it took me a long time to find out the real problem. With gradient descent my loss function was actually growing instead of getting smaller.
It turned out that my learning rate was too high. If you take too big of a step with gradient descent you can end up jumping over the minimum. And if you are really unlucky, like I was you end up jumping so far ahead that your error increases.
Lowering the learning rate should make the model converge. But it could take a long time.
Adam optimizer has momentum, that is, it doesn't just follow the instantaneous gradient, but it keeps track of the direction it was going before with a sort of velocity. This way, if you start going back and forth because of the gradient than the momentum will force you to go slower in this direction. This helps a lot! Adam has a few more tweeks other than momentum that make it the prefered deep learning optimizer.
If you want to read more about optimizers this blog post is very informative.
http://ruder.io/optimizing-gradient-descent/

Tensorflow: Linear regression with non-negative constraints

I am trying to implement a linear regression model in Tensorflow, with additional constraints (coming from the domain) that the W and b terms must be non-negative.
I believe there are a couple of ways to do this.
We can modify the cost function to penalize negative weights [Lagrangian approach] [See:TensorFlow - best way to implement weight constraints
We can compute the gradients ourselves and project them on [0, infinity] [Projected gradient approach]
Approach 1: Lagrangian
When I tried the first approach, I would often end up with negative b.
I had modified the cost function from:
cost = tf.reduce_sum(tf.pow(pred-Y, 2))/(2*n_samples)
to:
cost = tf.reduce_sum(tf.pow(pred-Y, 2))/(2*n_samples)
nn_w = tf.reduce_sum(tf.abs(W) - W)
nn_b = tf.reduce_sum(tf.abs(b) - b)
constraint = 100.0*nn_w + 100*nn_b
cost_with_constraint = cost + constraint
Keeping the coefficient of nn_b and nn_w to be very high leads to instability and very high cost.
Here is the complete code.
import numpy as np
import tensorflow as tf
n_samples = 50
train_X = np.linspace(1, 50, n_samples)
train_Y = 10*train_X + 6 +40*np.random.randn(50)
X = tf.placeholder("float")
Y = tf.placeholder("float")
# Set model weights
W = tf.Variable(np.random.randn(), name="weight")
b = tf.Variable(np.random.randn(), name="bias")
# Construct a linear model
pred = tf.add(tf.multiply(X, W), b)
# Gradient descent
learning_rate=0.0001
# Initializing the variables
init = tf.global_variables_initializer()
# Mean squared error
cost = tf.reduce_sum(tf.pow(pred-Y, 2))/(2*n_samples)
nn_w = tf.reduce_sum(tf.abs(W) - W)
nn_b = tf.reduce_sum(tf.abs(b) - b)
constraint = 1.0*nn_w + 100*nn_b
cost_with_constraint = cost + constraint
optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost_with_constraint)
training_epochs=200
with tf.Session() as sess:
sess.run(init)
# Fit all training data
cost_array = np.zeros(training_epochs)
W_array = np.zeros(training_epochs)
b_array = np.zeros(training_epochs)
for epoch in range(training_epochs):
for (x, y) in zip(train_X, train_Y):
sess.run(optimizer, feed_dict={X: x, Y: y})
W_array[epoch] = sess.run(W)
b_array[epoch] = sess.run(b)
cost_array[epoch] = sess.run(cost, feed_dict={X: train_X, Y: train_Y})
The following is the mean of b across 10 different runs.
0 -1.101268
1 0.169225
2 0.158363
3 0.706270
4 -0.371205
5 0.244424
6 1.312516
7 -0.069609
8 -1.032187
9 -1.711668
Clearly, the first approach is not optimal. Further, there is a lot of art involved in choosing the coefficient of penalty terms.
Approach 2: Projected gradient
I then thought to use the second approach, which is more guaranteed to work.
gr = tf.gradients(cost, [W, b])
We manually compute the gradients and update the W and b.
with tf.Session() as sess:
sess.run(init)
for epoch in range(training_epochs):
for (x, y) in zip(train_X, train_Y):
W_del, b_del = sess.run(gr, feed_dict={X: x, Y: y})
W = max(0, (W - W_del)*learning_rate) #Project the gradient on [0, infinity]
b = max(0, (b - b_del)*learning_rate) # Project the gradient on [0, infinity]
This approach seems to be very slow.
I am wondering if there is a better way to run the second approach, or guarantee the results with the first approach. Can we somehow allow the optimizer to ensure that the learnt weights are non-negative?
Edit: How to do this in Autograd
https://github.com/HIPS/autograd/issues/207

If you modify your linear model to:
pred = tf.add(tf.multiply(X, tf.abs(W)), tf.abs(b))
it will have the same effect as using only positive W and b values.
The reason your second approach is slow is that you clip the W and b values outside of the tensorflow graph. (Also it will not converge because (W - W_del)*learning_rate must instead be W - W_del*learning_rate)
edit:
You can implement the clipping using tensorflow graph like this:
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)
with tf.control_dependencies([train_step]):
clip_W = W.assign(tf.maximum(0., W))
clip_b = b.assign(tf.maximum(0., b))
train_step_with_clip = tf.group(clip_W, clip_b)
In this case W and b values will be clipped to 0 and not to small positive numbers.
Here is a small mnist example with clipping:
import tensorflow as tf
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x = tf.placeholder(tf.uint8, [None, 28, 28])
x_vec = tf.cast(tf.reshape(x, [-1, 784]), tf.float32) / 255.
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
y = tf.matmul(x_vec, W) + b
y_target = tf.placeholder(tf.uint8, [None])
y_target_one_hot = tf.one_hot(y_target, 10)
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_target_one_hot, logits=y))
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
with tf.control_dependencies([train_step]):
clip_W = W.assign(tf.maximum(0., W))
clip_b = b.assign(tf.maximum(0., b))
train_step_with_clip = tf.group(clip_W, clip_b)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_target_one_hot, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
with tf.Session() as sess:
tf.global_variables_initializer().run()
for i in range(1000):
sess.run(train_step_with_clip, feed_dict={
x: x_train[(i*100)%len(x_train):((i+1)*100)%len(x_train)],
y_target: y_train[(i*100)%len(x_train):((i+1)*100)%len(x_train)]})
if not i%100:
print("Min_W:", sess.run(tf.reduce_min(W)))
print("Min_b:", sess.run(tf.reduce_min(b)))
print("Accuracy:", sess.run(accuracy, feed_dict={
x: x_test,
y_target: y_test}))

I actually was not able to reproduce your problem of getting negative bs with your first approach.
But I do agree that this is not optimal for your use case and can result in negative values.
You should be able to constrain your parameters to non-negative values like so:
W *= tf.cast(W > 0., tf.float32)
b *= tf.cast(b > 0., tf.float32)
(exchange > with >= if necessary, the cast is necessary as the comparison operators will produce boolean values.
You then would optimize for the "standard cost" without the additional constraints.
However, this does not work in every case. For example, it should be avoided to initialize W or b with negative values in the beginning.
Your second (and probably better) approach can be accelerated by defining the update logic in the general computational graph, i.e. after the definition of cost
params = [W, b]
grads = tf.gradients(cost, params)
optimizer = [tf.assign(param, tf.maximum(0., param - grad*learning_rate))
for param, grad in zip(params, grads)]
I think your solution is slow because it creates new computation nodes every time which is probably very costly and repeated a lot inside the loops.
update using tensorflow optimizer
In my solution above not the gradients are clipped but rather the resulting update values.
Along the lines of this answer you could clip the gradients to be at most the value of the updated parameter like so:
params = [W, b]
opt = tf.train.GradientDescentOptimizer(learning_rate)
grads_and_vars = opt.compute_gradients(cost, params)
clipped_grads_vars = [(tf.clip_by_value(grad, -np.inf, var), var) for grad, var in grads_and_vars]
optimizer = opt.apply_gradients(clipped_grads_vars)
This way an update will never decrease a parameter to a value below 0.
However, I think this will not work in the case the updated variable is already negative.
Also, if the optimizing algorithm somehow multiplies the clipped gradient by a value greater than 1.
The latter might actually never happen, but I'm not 100% sure.

Getting very high values in linear regression

I am trying to make a simple MLP to predict values of a pixel of an image - original blog .
Here's my earlier attempt using Keras in python - link
I've tried to do the same in tensorflow, but I am getting very large output values (~10^12) when they should be less than 1.
Here's my code:
import numpy as np
import cv2
from random import shuffle
import tensorflow as tf
'''
Image preprocessing
'''
image_file = cv2.imread("Mona Lisa.jpg")
h = image_file.shape[0]
w = image_file.shape[1]
preX = []
preY = []
for i in xrange(h):
for j in xrange(w):
preX.append([i,j])
preY.append(image_file[i,j,:].astype('float32')/255.0)
print preX[:5], preY[:5]
zipped = [i for i in zip(preX,preY)]
shuffle(zipped)
X_train = np.array([i for (i,j) in zipped]).astype('float32')
Y_train = np.array([j for (i,j) in zipped]).astype('float32')
print X_train[:10], Y_train[:10]
'''
Tensorflow code
'''
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
x = tf.placeholder(tf.float32, shape=[None,2])
y = tf.placeholder(tf.float32, shape=[None,3])
'''
Layers
'''
w1 = weight_variable([2,300])
b1 = bias_variable([300])
L1 = tf.nn.relu(tf.matmul(X_train,w1)+b1)
w2 = weight_variable([300,3])
b2 = bias_variable([3])
y_model = tf.matmul(L1,w2)+b2
'''
Training
'''
# criterion
MSE = tf.reduce_mean(tf.square(tf.sub(y,y_model)))
# trainer
train_op = tf.train.GradientDescentOptimizer(learning_rate = 0.01).minimize(MSE)
nb_epochs = 10
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
cost = 0
for i in range(nb_epochs):
sess.run(train_op, feed_dict ={x: X_train, y: Y_train})
cost += sess.run(MSE, feed_dict ={x: X_train, y: Y_train})
cost /= nb_epochs
print cost
'''
Prediction
'''
pred = sess.run(y_model,feed_dict = {x:X_train})*255.0
print pred[:10]
output_image = []
index = 0
h = image_file.shape[0]
w = image_file.shape[1]
for i in xrange(h):
row = []
for j in xrange(w):
row.append(pred[index])
index += 1
row = np.array(row)
output_image.append(row)
output_image = np.array(output_image)
output_image = output_image.astype('uint8')
cv2.imwrite('out_mona_300x3_tf.png',output_image)

First of all, I think that instead of running the train_op and then the MSE
you can run both ops in a list and reduce your computational cost significantly.
for i in range(nb_epochs):
cost += sess.run([MSE, train_op], feed_dict ={x: X_train, y: Y_train})
Secondly, I suggest always writing out your cost function so you can see what is going on during the training phase. Either manually print it out or use tensorboard to log your cost and plot it (you can find examples on the official tf page).
You can also monitor your weights to see that they aren't blowing up.
A few things you can try:
Reduce learning rate, add regularization to weights.
Check that your training set (pixels) really consist of the values that
you expect them to.

You give the input layer weights and the output layer weights the same names w and b, so it seems something goes wrong in the gradient-descent procedure. Actually I'm surprised tensorflow doesn't issue an error or at leas a warning (or am I missing something?)