I need to save some values to specific places in a tensorflow array:
import tensorflow as tf
import numpy as np
AVG = tf.Variable([0, 0, 0, 0, 0], name='data')
for i in range(5):
data = np.random.randint(1000, size=10000)
AVG += np.average(data)
I need to make it average each iteration in different places of the AVG variable. Is this doable ?
You can use tf.scatter_add. Here is a complete working program :
import tensorflow as tf
import numpy as np
AVG = tf.Variable([0, 0, 0, 0, 0], name='data')
for i in range(5):
data = np.random.randint(1000, size=10000)
AVG = tf.scatter_add(AVG, [i], [np.average(data).astype('int')])
sess = tf.InteractiveSession()
sess.run(tf.initialize_all_variables())
print(AVG.eval())
Related
My prediction is y_hat = [ 0.57,0.05,0.14,0.10,0.14] and target is
target =[ 1, 0, 0, 0, 0 ].
I need to calculate Cross Entropy loss by NumPy and Pytorch loss function.
Using NumPy my formula is -np.sum(target*np.log(y_hat)), and I got 0.5621189181535413
However, using Pytorch:
loss = nn.CrossEntropyLoss()
output = torch.FloatTensor([0.57,0.05,0.14,0.10,0.14])
label = torch.FloatTensor([1,0,0,0,0])
loss_value = loss(output, label)
print(loss_value)
Gives tensor(1.2586), which is different.
You need to apply the softmax function to your y_hat vector before computing cross-entropy loss. For example, you can use scipy.special.softmax().
>>> from scipy.special import softmax
>>> import numpy as np
>>> y_hat = [0.57, 0.05, 0.14, 0.10, 0.14]
>>> target =[1, 0, 0, 0, 0]
>>> y_hat = softmax(y_hat)
>>> -np.sum(target * np.log(y_hat))
1.2586146726011722
Which agrees with the result from Pytorch.
Can you modify the value of a specific tensor?
For example:
x = tf.zeros((2, 2))
x[0, 0] = 1 # pseudo-code
print(x) # <Tensor ... numpy=[[1, 0], [0, 0]]>
You can but you need to set the tensor to be a Variable and not a constant.
import tensorflow as tf
import numpy as np
x = np.zeros((2,2))
x_var = tf.Variable(x)
x[0,0]=1
tf.assign(x_var ,x)
I am doing a task in which i have data in form of frames stored in batches. Dimension of a batch is like (batch_size,400), i want to find index of first occurrence of number 1 in each 400 length frame.
currently i m using for loop over batch size but since data is very larger it is very time consuming
Any other Efficient method using some matrix operation in tensorflow or numpy would
In TensorFlow:
import tensorflow as tf
def index_of_first_tf(batch, value):
eq = tf.equal(batch, value)
has_value = tf.reduce_any(eq, axis=-1)
_, idx = tf.math.top_k(tf.cast(eq, tf.int8))
idx = tf.squeeze(idx, -1)
return tf.where(has_value, idx, -tf.ones_like(idx))
In NumPy:
import numpy as np
def index_of_first_np(batch, value):
eq = np.equal(batch, value)
has_value = np.any(eq, axis=-1)
idx = np.argmax(eq, axis=-1)
idx[~has_value] = -1
return idx
Tests:
import tensorflow as tf
batch = [[0, 1, 2, 3],
[1, 2, 1, 0],
[0, 2, 3, 4]]
value = 1
print(index_of_first_np(batch, value))
# [ 1 0 -1]
with tf.Graph().as_default(), tf.Session() as sess:
print(sess.run(index_of_first_tf(batch, value)))
# [ 1 0 -1]
I am very new to the deep learning and computer vision. I want to do some face recognition project. For that I downloaded some images from Internet and converted to Tensorflow dataset by the help of this article from tensorflow documentation. Now I want to convert that dataset to pandas dataframe in order to convert that to csv files. I tried a lot but am unable to do it.
Can someone help me with it.
Here is the code for making datasets and and then some of the wrong code which I tried for this.
import tensorflow as tf
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
filenames = tf.constant(['al.jpg', 'al2.jpg', 'al3.jpg', 'al4.jpeg','al5.jpeg', 'al6.jpeg','al7.jpg','al8.jpeg', '5.jpg', 'hrit8.jpeg', 'Hrithik-Roshan.jpg', 'Hrithik.jpg', 'hriti1.jpeg', 'hriti2.jpg', 'hriti3.jpeg', 'hritik4.jpeg', 'hritik5.jpg', 'hritk9.jpeg', 'index.jpeg', 'sah.jpeg', 'sah1.jpeg', 'sah3.jpeg', 'sah4.jpg', 'sah5.jpg','sah6.jpg','sah7.jpg'])
labels = tf.constant([1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 2, 2, 2, 2, 2, 2])
dataset = tf.data.Dataset.from_tensor_slices((filenames, labels))
def _parse_function(filename, label):
image_string = tf.read_file(filename)
image_decoded = tf.image.decode_jpeg(image_string,channels=3)
image_resized = tf.image.resize_images(image_decoded, [28, 28])
return image_resized, label
dataset = dataset.map(_parse_function)
dataset = dataset.shuffle(buffer_size=100)
dataset = dataset.batch(26)
iterator = dataset.make_one_shot_iterator()
image,labels = iterator.get_next()
sess = tf.Session()
print(sess.run([image, labels]))
Initially I just tried to use df = pd.DataFrame(dataset)
Then i got following error:
enter code here
ValueError Traceback (most recent call last)
<ipython-input-15-d5503ae4603d> in <module>()
----> 1 df = pd.DataFrame((dataset))
~/anaconda3/lib/python3.6/site-packages/pandas/core/frame.py in __init__(self, data, index, columns, dtype, copy)
402 dtype=values.dtype, copy=False)
403 else:
--> 404 raise ValueError('DataFrame constructor not properly called!')
405
406 NDFrame.__init__(self, mgr, fastpath=True)
ValueError: DataFrame constructor not properly called!
Thereafter I came across this article I got my mistake that in tensorflow anything exist only within a session. So I tried following code:
with tf.Session() as sess:
df = pd.DataFrame(sess.run(dataset))
Please pardon me if i did stupidest mistake because i wrote above code from this analogy print(sess.run(dataset)) and got a much bigger error:
TypeError: Fetch argument <BatchDataset shapes: ((?, 28, 28, 3), (?,)), types: (tf.float32, tf.int32)> has invalid type <class 'tensorflow.python.data.ops.dataset_ops.BatchDataset'>, must be a string or Tensor. (Can not convert a BatchDataset into a Tensor or Operation.)
I think you could use map like this. I assumed that you want to add a numpy array to a data frame as described here. But you have to append one by one and also figure out how this whole array fits in one column of the data frame.
import tensorflow as tf
import pandas as pd
filenames = tf.constant(['C:/Machine Learning/sunflower/50987813_7484bfbcdf.jpg'])
labels = tf.constant([1])
dataset = tf.data.Dataset.from_tensor_slices((filenames, labels))
sess = tf.Session()
def convert_to_dataframe(filename, label):
print ( pd.DataFrame.from_records(filename))
return filename, label
def _parse_function(filename, label):
image_string = tf.read_file(filename)
image_decoded = tf.image.decode_jpeg(image_string,channels=3)
image_resized = tf.image.resize_images(image_decoded, [28, 28])
return image_resized, label
dataset = dataset.map(_parse_function)
dataset = dataset.map( lambda filename, label: tf.py_func(convert_to_dataframe,
[filename, label],
[tf.float32,tf.int32]))
dataset = dataset.shuffle(buffer_size=100)
dataset = dataset.batch(26)
iterator = dataset.make_one_shot_iterator()
image,labels = iterator.get_next()
sess.run([image, labels])
One easy way to do it is to save the dataset into normal csv file, and then directly read the csv file into pandas dataframe.
import tensorflow_datasets as tfds
# Construct a tf.data.Dataset
ds = tfds.load('civil_comments/CivilCommentsCovert', split='train')
#read the dataset into a tensorflow styled_dataframe
df = tfds.as_dataframe(ds)
#save the dataframe into csv file
df.to_csv("/.../.../Desktop/covert_toxicity.csv")
#read the csv file as normal, then you have the df you need
import pandas as pd
file_path = "/.../.../Desktop/covert_toxicity.csv"
df = pd.read_csv(file_path, header = 0, sep=",")
df
A more simpler way to convert a TensorFlow object to a dataframe would be to convert the TensorFlow object to a numpy array and pass the pandas DataFrame class.
import pandas as pd
dataset = pd.DataFrame(labels.numpy(), columns=filenames)
There is a function in numpy that inserts given values to the array:
https://docs.scipy.org/doc/numpy/reference/generated/numpy.insert.html
Is there something similar in tensorflow?
Alternatively, is there a function in tensorflow that can do tensor upsampling using zeros in between values of a tensor?
tf.nn.conv2d_transpose can do this upsampling (with careful design of output_shape and strides). A sample code:
import tensorflow as tf
import numpy as np
input = tf.convert_to_tensor(np.ones((1, 20, 20, 1)))
input = tf.cast(input, tf.float32)
b = np.zeros((3, 3, 1, 1))
b[1, 1, 0, 0] = 1
weight = tf.convert_to_tensor(b)
weight = tf.cast(weight, tf.float32)
output = tf.nn.conv2d_transpose(input, weight, output_shape=(1, 40, 40, 1), strides=[1, 2, 2, 1])
sess = tf.Session()
print sess.run(output[0, :, :, 0])
I believe checking its api will help you more.