I am basically clustering some of my documents using mini_batch_kmeans and kmeans algorithm. I simply followed the tutorial is the scikit-learn website the link for that is given below:
http://scikit-learn.org/stable/auto_examples/text/document_clustering.html
They are using some of the method for the vectorizing one of which is HashingVectorizer. In the hashingVectorizer they are making a pipeline with TfidfTransformer() method.
# Perform an IDF normalization on the output of HashingVectorizer
hasher = HashingVectorizer(n_features=opts.n_features,
stop_words='english', non_negative=True,
norm=None, binary=False)
vectorizer = make_pipeline(hasher, TfidfTransformer())
Once doing so, the vectorizer what I get from that does not have the method get_feature_names(). But since I am using it for clustering, I need to get the "terms" using this "get_feature_names()"
terms = vectorizer.get_feature_names()
for i in range(true_k):
print("Cluster %d:" % i, end='')
for ind in order_centroids[i, :10]:
print(' %s' % terms[ind], end='')
print()
How do I solve this error?
My whole code is show below:
X_train_vecs, vectorizer = vector_bow.count_tfidf_vectorizer(_contents)
mini_kmeans_batch = MiniBatchKmeansTechnique()
# MiniBatchKmeans without the LSA dimensionality reduction
mini_kmeans_batch.mini_kmeans_technique(number_cluster=8, X_train_vecs=X_train_vecs,
vectorizer=vectorizer, filenames=_filenames, contents=_contents, is_dimension_reduced=False)
The count vectorizor piped with tfidf.
def count_tfidf_vectorizer(self,contents):
count_vect = CountVectorizer()
vectorizer = make_pipeline(count_vect,TfidfTransformer())
X_train_vecs = vectorizer.fit_transform(contents)
print("The count of bow : ", X_train_vecs.shape)
return X_train_vecs, vectorizer
and the mini_batch_kmeans class is as below:
class MiniBatchKmeansTechnique():
def mini_kmeans_technique(self, number_cluster, X_train_vecs, vectorizer,
filenames, contents, svd=None, is_dimension_reduced=True):
km = MiniBatchKMeans(n_clusters=number_cluster, init='k-means++', max_iter=100, n_init=10,
init_size=1000, batch_size=1000, verbose=True, random_state=42)
print("Clustering sparse data with %s" % km)
t0 = time()
km.fit(X_train_vecs)
print("done in %0.3fs" % (time() - t0))
print()
cluster_labels = km.labels_.tolist()
print("List of the cluster names is : ",cluster_labels)
data = {'filename':filenames, 'contents':contents, 'cluster_label':cluster_labels}
frame = pd.DataFrame(data=data, index=[cluster_labels], columns=['filename', 'contents', 'cluster_label'])
print(frame['cluster_label'].value_counts(sort=True,ascending=False))
print()
grouped = frame['cluster_label'].groupby(frame['cluster_label'])
print(grouped.mean())
print()
print("Top Terms Per Cluster :")
if is_dimension_reduced:
if svd != None:
original_space_centroids = svd.inverse_transform(km.cluster_centers_)
order_centroids = original_space_centroids.argsort()[:, ::-1]
else:
order_centroids = km.cluster_centers_.argsort()[:, ::-1]
terms = vectorizer.get_feature_names()
for i in range(number_cluster):
print("Cluster %d:" % i, end=' ')
for ind in order_centroids[i, :10]:
print(' %s' % terms[ind], end=',')
print()
print("Cluster %d filenames:" % i, end='')
for file in frame.ix[i]['filename'].values.tolist():
print(' %s,' % file, end='')
print()
Pipeline doesn't have get_feature_names() method, as it is not straightforward to implement this method for Pipeline - one needs to consider all pipeline steps to get feature names. See https://github.com/scikit-learn/scikit-learn/issues/6424, https://github.com/scikit-learn/scikit-learn/issues/6425, etc. - there is a lot of related tickets and several attempts to fix it.
If your pipeline is simple (TfidfVectorizer followed by MiniBatchKMeans) then you can get feature names from TfidfVectorizer.
If you want to use HashingVectorizer, it is more complicated, as HashingVectorizer doesn't provide feature names by design. HashingVectorizer doesn't store vocabulary, and uses hashes instead - it means it can be applied in online setting, and that it dosn't require any RAM - but the tradeoff is exactly that you don't get feature names.
It is still possible to get feature names from HashingVectorizer though; to do this you need to apply it for a sample of documents, store which hashes correspond to which words, and this way learn what these hashes mean, i.e. what are the feature names. There may be collisions, so it is not possible to be 100% sure the feature name is correct, but usually this approach works ok. This approach is implemented in eli5 library; see http://eli5.readthedocs.io/en/latest/tutorials/sklearn-text.html#debugging-hashingvectorizer for an example. You will have to do something like this, using InvertableHashingVectorizer:
from eli5.sklearn import InvertableHashingVectorizer
ivec = InvertableHashingVectorizer(vec) # vec is a HashingVectorizer instance
# X_sample is a sample from contents; you can use the
# whole contents array, or just e.g. every 10th element
ivec.fit(content_sample)
hashing_feat_names = ivec.get_feature_names()
Then you can use hashing_feat_names as your feature names, as TfidfTransformer doesn't change input vector size and just scales the same features.
From the make_pipeline documentation:
This is a shorthand for the Pipeline constructor; it does not require, and
does not permit, naming the estimators. Instead, their names will be set
to the lowercase of their types automatically.
so, in order to access the feature names, after you have fitted to data, you can:
# Perform an IDF normalization on the output of HashingVectorizer
from sklearn.feature_extraction.text import HashingVectorizer, TfidfVectorizer
from sklearn.pipeline import make_pipeline
hasher = HashingVectorizer(n_features=10,
stop_words='english', non_negative=True,
norm=None, binary=False)
tfidf = TfidfVectorizer()
vectorizer = make_pipeline(hasher, tfidf)
# ...
# fit to the data
# ...
# use the instance's class name to lower
terms = vectorizer.named_steps[tfidf.__class__.__name__.lower()].get_feature_names()
# or to be more precise, as used in `_name_estimators`:
# terms = vectorizer.named_steps[type(tfidf).__name__.lower()].get_feature_names()
# btw TfidfTransformer and HashingVectorizer do not have get_feature_names afaik
Hope this helps, good luck!
Edit: After seeing your updated question with the example you follow, #Vivek Kumar is correct, this code terms = vectorizer.get_feature_names() will not run for the pipeline but only when:
vectorizer = TfidfVectorizer(max_df=0.5, max_features=opts.n_features,
min_df=2, stop_words='english',
use_idf=opts.use_idf)
Related
I would like that every client train his model with a function pretrainthat I wrote below :
def pretrain(model):
resnet_output = model.output
layer1 = tf.keras.layers.GlobalAveragePooling2D()(resnet_output)
layer2 = tf.keras.layers.Dense(units=zdim*2, activation='relu')(layer1)
model_output = tf.keras.layers.Dense(units=zdim)(layer2)
model = tf.keras.Model(model.input, model_output)
iterations_per_epoch = determine_iterations_per_epoch()
total_iterations = iterations_per_epoch*num_epochs
optimizer = tf.keras.optimizers.SGD(learning_rate=learning_rate, momentum=0.9)
checkpoint = tf.train.Checkpoint(step=tf.Variable(1), optimizer=optimizer, net=model)
manager = tf.train.CheckpointManager(checkpoint, pretrain_save_path, max_to_keep=10)
current_epoch = tf.cast(tf.floor(optimizer.iterations/iterations_per_epoch), tf.int64)
batch = client_data(0)
batch = client_data(0).batch(2)
epoch_loss = []
for (image1, image2) in batch:
loss, gradients = train_step(model, image1, image2)
epoch_loss.append(loss)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
# if tf.reduce_all(tf.equal(epoch, current_epoch+1)):
print("Loss after epoch {}: {}".format(current_epoch, sum(epoch_loss)/len(epoch_loss)))
#print("Learning rate: {}".format(learning_rate(optimizer.iterations)))
epoch_loss = []
current_epoch += 1
if current_epoch % 50 == 0:
save_path = manager.save()
print("Saved model for epoch {}: {}".format(current_epoch, save_path))
save_path = manager.save()
model.save("model.h5")
model.save_weights("saved_weights.h5")
But as we know that TFF has a predefined function :
iterative_process = tff.learning.build_federated_averaging_process(...)
So please, how can I proceed ? Thanks
There are a few ways that one could proceed along similar lines.
First it is important to note that TFF is functional--one can use things like writing to / reading from files to manage state (as TF allows this), but it is not in the interface TFF exposes to users--while something involving writing to / reading from a file (IE, manipulating state without passing it through function parameters and results), this should at best be considered an implementation detail. It's something that TFF does not encourage.
By slightly refactoring your code above, however, I think this kind of application can fit quite nicely in TFF's programming model. We will want to define something like:
#tff.tf_computation
#tf.function
def pretrain_client_model(model, client_dataset):
# perhaps do dataset processing you want...
for batch in client_dataset:
# do model training
return model.weights() # or some tensor structure representing the trained model weights
Once your implementation looks something like this, you will be able to wire it in to a custom iterative process. The canned function you mention (build_federated_averaging_process) really just constructs an instance of tff.templates.IterativeProcess; you are always, however, free to write your own instance of this class.
Several tutorials take us through this process, this probably being the simplest. For a finished code example of a standalone iterative process implementation, see simple_fedavg.py.
Would anyone know how to implement the FedProx optimisation algorithm with TensorFlow Federated? The only implementation that seems to be available online was developed directly with TensorFlow. A TFF implementation would enable an easier comparison with experiments that utilise FedAvg which the framework supports.
This is the link to the FedProx repo: https://github.com/litian96/FedProx
Link to the paper: https://arxiv.org/abs/1812.06127
At this moment, FedProx implementation is not available. I agree it would be a valuable algorithm to have.
If you are interested in contributing FedProx, the best place to start would be simple_fedavg which is a minimal implementation of FedAvg meant as a starting point for extensions -- see the readme there for more details.
I think the major change would need to happen to the client_update method, where you would add the proximal term depending on model_weights and initial_weights to the loss computed in forward pass.
I provide below my implementation of FedProx in TFF. I am not 100% sure that this is the right implementation; I post this answer also for discussing on actual code example.
I tried to follow the suggestions in the Jacub Konecny's answer and comment.
Starting from the simple_fedavg (referring to the TFF Github repo), I just modified the client_update method, and specifically changing the input argument for calculating the gradient with the GradientTape, i.e. instaead of just passing in input the outputs.loss, the tape calculates the gradient considering the outputs.loss + proximal_term previosuly (and iteratively) calculated.
#tf.function
def client_update(model, dataset, server_message, client_optimizer):
"""Performans client local training of "model" on "dataset".Args:
model: A "tff.learning.Model".
dataset: A "tf.data.Dataset".
server_message: A "BroadcastMessage" from server.
client_optimizer: A "tf.keras.optimizers.Optimizer".
Returns:
A "ClientOutput".
"""
def difference_model_norm_2_square(global_model, local_model):
"""Calculates the squared l2 norm of a model difference (i.e.
local_model - global_model)
Args:
global_model: the model broadcast by the server
local_model: the current, in-training model
Returns: the squared norm
"""
model_difference = tf.nest.map_structure(lambda a, b: a - b,
local_model,
global_model)
squared_norm = tf.square(tf.linalg.global_norm(model_difference))
return squared_norm
model_weights = model.weights
initial_weights = server_message.model_weights
tf.nest.map_structure(lambda v, t: v.assign(t), model_weights,
initial_weights)
num_examples = tf.constant(0, dtype=tf.int32)
loss_sum = tf.constant(0, dtype=tf.float32)
# Explicit use `iter` for dataset is a trick that makes TFF more robust in
# GPU simulation and slightly more performant in the unconventional usage
# of large number of small datasets.
for batch in iter(dataset):
with tf.GradientTape() as tape:
outputs = model.forward_pass(batch)
# ------ FedProx ------
mu = tf.constant(0.2, dtype=tf.float32)
prox_term =(mu/2)*difference_model_norm_2_square(model_weights.trainable, initial_weights.trainable)
fedprox_loss = outputs.loss + prox_term
# Letting GradientTape dealing with the FedProx's loss
grads = tape.gradient(fedprox_loss, model_weights.trainable)
client_optimizer.apply_gradients(zip(grads, model_weights.trainable))
batch_size = tf.shape(batch['x'])[0]
num_examples += batch_size
loss_sum += outputs.loss * tf.cast(batch_size, tf.float32)
weights_delta = tf.nest.map_structure(lambda a, b: a - b,
model_weights.trainable,
initial_weights.trainable)
client_weight = tf.cast(num_examples, tf.float32)
return ClientOutput(weights_delta, client_weight, loss_sum / client_weight)
I want to fit a Siamese CNN with multiple inputs that are stored in my memory and no label (just an arbitrary dummy label). Therefore, I had to write my own data_generator function for using a CNN model in Keras.
My data generator is of the following form
class DataGenerator(keras.utils.Sequence):
def __init__(self, train_data, train_triplets, batch_size=32, dim=(128,128), n_channels=3, shuffle=True):
self.dim = dim
self.batch_size = batch_size
#Added
self.train_data = train_data
self.train_triplets = train_triplets
self.n_channels = n_channels
self.shuffle = shuffle
self.on_epoch_end()
def __len__(self):
'Denotes the number of batches per epoch'
n_row = self.train_triplets.shape[0]
return int(np.floor(n_row / self.batch_size))
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
#print(index)
indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]
# Find list of IDs
list_IDs_temp = self.train_triplets.iloc[indexes,]
# Generate data
[anchor, positive, negative] = self.__data_generation(list_IDs_temp)
y_train = np.random.randint(2, size=(1,2,self.batch_size)).T
return [anchor,positive, negative], y_train
def on_epoch_end(self):
'Updates indexes after each epoch'
n_row = self.train_triplets.shape[0]
self.indexes = np.arange(n_row)
if self.shuffle == True:
np.random.shuffle(self.indexes)
def __data_generation(self, list_IDs_temp):
'Generates data containing batch_size samples'
# anchor positive and negatives: (n_samples, *dim, n_channels)
# Initialization
anchor = np.zeros((self.batch_size,*self.dim,self.n_channels))
positive = np.zeros((self.batch_size,*self.dim,self.n_channels))
negative = np.zeros((self.batch_size,*self.dim,self.n_channels))
nrow_temp = list_IDs_temp.shape[0]
# Generate data
for i in range(nrow_temp):
list_ind = list_IDs_temp.iloc[i,]
anchor[i] = self.train_data[list_ind[0]]
positive[i] = self.train_data[list_ind[1]]
negative[i] = self.train_data[list_ind[2]]
return [anchor, positive, negative]
where train_data is a list of all images and train triplets a data frame containing image indices to create my inputs containing of a triplet of images.
Now, I want to do some data augmenting for each mini batch supplied to my CNN. I have tried to integrate the ImageDataGenarator of Keras but I couldn't implement it in my code. Is it somehow possible to do it ? I am not very experienced with python and would appreciate any help.
Does this article answer your question?
To put it in a nutshell, Kera's ImageDataGenerator lacks flexibility when it comes to personalized batch generators, and the easiest way to still use data augmentation is simply to switch to another data augmentation tool (like the albumentations library described in the previous article, but you could also use imgaug as well).
I just want to warn you that I encountered several issues with albumentations (that I described in this question on GitHub, but for now I still have had no answers), so maybe using imgaug is a better idea.
Hope this helps, good luck with your model !
from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
Hello I have the following list of comments:
comments = ['I am very agry','this is not interesting','I am very happy']
These are the corresponding labels:
sents = ['angry','indiferent','happy']
I am using tfidf to vectorize these comments as follows:
tfidf_vectorizer = TfidfVectorizer(analyzer='word')
tfidf = tfidf_vectorizer.fit_transform(comments)
from sklearn import preprocessing
I am using label encoder to vectorize the labels:
le = preprocessing.LabelEncoder()
le.fit(sents)
labels = le.transform(sents)
print(labels.shape)
from sklearn.linear_model import PassiveAggressiveClassifier
from sklearn.model_selection import train_test_split
with open('tfidf.pickle','wb') as idxf:
pickle.dump(tfidf, idxf, pickle.HIGHEST_PROTOCOL)
with open('tfidf_vectorizer.pickle','wb') as idxf:
pickle.dump(tfidf_vectorizer, idxf, pickle.HIGHEST_PROTOCOL)
Here I am using passive aggressive to fit the model:
clf2 = PassiveAggressiveClassifier()
with open('passive.pickle','wb') as idxf:
pickle.dump(clf2, idxf, pickle.HIGHEST_PROTOCOL)
with open('passive.pickle', 'rb') as infile:
clf2 = pickle.load(infile)
with open('tfidf_vectorizer.pickle', 'rb') as infile:
tfidf_vectorizer = pickle.load(infile)
with open('tfidf.pickle', 'rb') as infile:
tfidf = pickle.load(infile)
Here I am trying to test the usage of partial fit as follows with three new comments and their corresponding labels:
new_comments = ['I love the life','I hate you','this is not important']
new_labels = [1,0,2]
vec_new_comments = tfidf_vectorizer.transform(new_comments)
print(clf2.predict(vec_new_comments))
clf2.partial_fit(vec_new_comments, new_labels)
The problem is that I am not getting the right results after the partial fit as follows:
print('AFTER THIS UPDATE THE RESULT SHOULD BE 1,0,2??')
print(clf2.predict(vec_new_comments))
however I am getting this output:
[2 2 2]
So I really appreciate support to find, why the model is not being updated if I am testing it with the same examples that it has used to be trained the desired output should be:
[1,0,2]
I would like to appreciate support to ajust maybe the hyperparameters to see the desired output.
this is the complete code, to show the partial fit:
from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
import sys
from sklearn.metrics.pairwise import cosine_similarity
import random
comments = ['I am very agry','this is not interesting','I am very happy']
sents = ['angry','indiferent','happy']
tfidf_vectorizer = TfidfVectorizer(analyzer='word')
tfidf = tfidf_vectorizer.fit_transform(comments)
#print(tfidf.shape)
from sklearn import preprocessing
le = preprocessing.LabelEncoder()
le.fit(sents)
labels = le.transform(sents)
from sklearn.linear_model import PassiveAggressiveClassifier
from sklearn.model_selection import train_test_split
with open('tfidf.pickle','wb') as idxf:
pickle.dump(tfidf, idxf, pickle.HIGHEST_PROTOCOL)
with open('tfidf_vectorizer.pickle','wb') as idxf:
pickle.dump(tfidf_vectorizer, idxf, pickle.HIGHEST_PROTOCOL)
clf2 = PassiveAggressiveClassifier()
clf2.fit(tfidf, labels)
with open('passive.pickle','wb') as idxf:
pickle.dump(clf2, idxf, pickle.HIGHEST_PROTOCOL)
with open('passive.pickle', 'rb') as infile:
clf2 = pickle.load(infile)
with open('tfidf_vectorizer.pickle', 'rb') as infile:
tfidf_vectorizer = pickle.load(infile)
with open('tfidf.pickle', 'rb') as infile:
tfidf = pickle.load(infile)
new_comments = ['I love the life','I hate you','this is not important']
new_labels = [1,0,2]
vec_new_comments = tfidf_vectorizer.transform(new_comments)
clf2.partial_fit(vec_new_comments, new_labels)
print('AFTER THIS UPDATE THE RESULT SHOULD BE 1,0,2??')
print(clf2.predict(vec_new_comments))
However I got:
AFTER THIS UPDATE THE RESULT SHOULD BE 1,0,2??
[2 2 2]
Well there are multiple problems with your code. I will start by stating the obvious ones to more complex ones:
You are pickling the clf2 before it has learnt anything. (ie. you pickle it as soon as it is defined, it doesnt serve any purpose). If you are only testing, then fine. Otherwise they should be pickled after the fit() or equivalent calls.
You are calling clf2.fit() before the clf2.partial_fit(). This defeats the whole purpose of partial_fit(). When you call fit(), you essentially fix the classes (labels) that the model will learn about. In your case it is acceptable, because on your subsequent call to partial_fit() you are giving the same labels. But still it is not a good practice.
See this for more details
In a partial_fit() scenario, dont call the fit() ever. Always call the partial_fit() with your starting data and new coming data. But make sure that you supply all the labels you want the model to learn in the first call to parital_fit() in a parameter classes.
Now the last part, about your tfidf_vectorizer. You call fit_transform()(which is essentially fit() and then transformed() combined) on tfidf_vectorizer with comments array. That means that it on subsequent calls to transform() (as you did in transform(new_comments)), it will not learn new words from new_comments, but only use the words which it saw during the call to fit()(words present in comments).
Same goes for LabelEncoder and sents.
This again is not prefereble in a online learning scenario. You should fit all the available data at once. But since you are trying to use the partial_fit(), we assume that you have very large dataset which may not fit in memory at once. So you would want to apply some sort of partial_fit to TfidfVectorizer as well. But TfidfVectorizer doesnt support partial_fit(). In fact its not made for large data. So you need to change your approach. See the following questions for more details:-
Updating the feature names into scikit TFIdfVectorizer
How can i reduce memory usage of Scikit-Learn Vectorizers?
All things aside, if you change just the tfidf part of fitting the whole data (comments and new_comments at once), you will get your desired results.
See the below code changes (I may have organized it a bit and renamed vec_new_comments to new_tfidf, please go through it with attention):
comments = ['I am very agry','this is not interesting','I am very happy']
sents = ['angry','indiferent','happy']
new_comments = ['I love the life','I hate you','this is not important']
new_sents = ['happy','angry','indiferent']
tfidf_vectorizer = TfidfVectorizer(analyzer='word')
le = preprocessing.LabelEncoder()
# The below lines are important
# I have given the whole data to fit in tfidf_vectorizer
tfidf_vectorizer.fit(comments + new_comments)
# same for `sents`, but since the labels dont change, it doesnt matter which you use, because it will be same
# le.fit(sents)
le.fit(sents + new_sents)
Below is the Not so preferred code (which you are using, and about which I talked in point 2), but results are good as long as you make the above changes.
tfidf = tfidf_vectorizer.transform(comments)
labels = le.transform(sents)
clf2.fit(tfidf, labels)
print(clf2.predict(tfidf))
# [0 2 1]
new_tfidf = tfidf_vectorizer.transform(new_comments)
new_labels = le.transform(new_sents)
clf2.partial_fit(new_tfidf, new_labels)
print(clf2.predict(new_tfidf))
# [1 0 2] As you wanted
Correct approach, or the way partial_fit() is intended to be used:
# Declare all labels that you want the model to learn
# Using classes learnt by labelEncoder for this
# In any calls to `partial_fit()`, all labels should be from this array only
all_classes = le.transform(le.classes_)
# Notice the parameter classes here
# It needs to present first time
clf2.partial_fit(tfidf, labels, classes=all_classes)
print(clf2.predict(tfidf))
# [0 2 1]
# classes is not present here
clf2.partial_fit(new_tfidf, new_labels)
print(clf2.predict(new_tfidf))
# [1 0 2]
Is there a way to plot both the training losses and validation losses on the same graph?
It's easy to have two separate scalar summaries for each of them individually, but this puts them on separate graphs. If both are displayed in the same graph it's much easier to see the gap between them and whether or not they have begin to diverge due to overfitting.
Is there a built in way to do this? If not, a work around way? Thank you much!
The work-around I have been doing is to use two SummaryWriter with different log dir for training set and cross-validation set respectively. And you will see something like this:
Rather than displaying the two lines separately, you can instead plot the difference between validation and training losses as its own scalar summary to track the divergence.
This doesn't give as much information on a single plot (compared with adding two summaries), but it helps with being able to compare multiple runs (and not adding multiple summaries per run).
Just for anyone coming accross this via a search: The current best practice to achieve this goal is to just use the SummaryWriter.add_scalars method from torch.utils.tensorboard. From the docs:
from torch.utils.tensorboard import SummaryWriter
writer = SummaryWriter()
r = 5
for i in range(100):
writer.add_scalars('run_14h', {'xsinx':i*np.sin(i/r),
'xcosx':i*np.cos(i/r),
'tanx': np.tan(i/r)}, i)
writer.close()
# This call adds three values to the same scalar plot with the tag
# 'run_14h' in TensorBoard's scalar section.
Expected result:
Many thanks to niko for the tip on Custom Scalars.
I was confused by the official custom_scalar_demo.py because there's so much going on, and I had to study it for quite a while before I figured out how it worked.
To show exactly what needs to be done to create a custom scalar graph for an existing model, I put together the following complete example:
# + <
# We need these to make a custom protocol buffer to display custom scalars.
# See https://developers.google.com/protocol-buffers/
from tensorboard.plugins.custom_scalar import layout_pb2
from tensorboard.summary.v1 import custom_scalar_pb
# >
import tensorflow as tf
from time import time
import re
# Initial values
(x0, y0) = (-1, 1)
# This is useful only when re-running code (e.g. Jupyter).
tf.reset_default_graph()
# Set up variables.
x = tf.Variable(x0, name="X", dtype=tf.float64)
y = tf.Variable(y0, name="Y", dtype=tf.float64)
# Define loss function and give it a name.
loss = tf.square(x - 3*y) + tf.square(x+y)
loss = tf.identity(loss, name='my_loss')
# Define the op for performing gradient descent.
minimize_step_op = tf.train.GradientDescentOptimizer(0.092).minimize(loss)
# List quantities to summarize in a dictionary
# with (key, value) = (name, Tensor).
to_summarize = dict(
X = x,
Y_plus_2 = y + 2,
)
# Build scalar summaries corresponding to to_summarize.
# This should be done in a separate name scope to avoid name collisions
# between summaries and their respective tensors. The name scope also
# gives a title to a group of scalars in TensorBoard.
with tf.name_scope('scalar_summaries'):
my_var_summary_op = tf.summary.merge(
[tf.summary.scalar(name, var)
for name, var in to_summarize.items()
]
)
# + <
# This constructs the layout for the custom scalar, and specifies
# which scalars to plot.
layout_summary = custom_scalar_pb(
layout_pb2.Layout(category=[
layout_pb2.Category(
title='Custom scalar summary group',
chart=[
layout_pb2.Chart(
title='Custom scalar summary chart',
multiline=layout_pb2.MultilineChartContent(
# regex to select only summaries which
# are in "scalar_summaries" name scope:
tag=[r'^scalar_summaries\/']
)
)
])
])
)
# >
# Create session.
with tf.Session() as sess:
# Initialize session.
sess.run(tf.global_variables_initializer())
# Create writer.
with tf.summary.FileWriter(f'./logs/session_{int(time())}') as writer:
# Write the session graph.
writer.add_graph(sess.graph) # (not necessary for scalars)
# + <
# Define the layout for creating custom scalars in terms
# of the scalars.
writer.add_summary(layout_summary)
# >
# Main iteration loop.
for i in range(50):
current_summary = sess.run(my_var_summary_op)
writer.add_summary(current_summary, global_step=i)
writer.flush()
sess.run(minimize_step_op)
The above consists of an "original model" augmented by three blocks of code indicated by
# + <
[code to add custom scalars goes here]
# >
My "original model" has these scalars:
and this graph:
My modified model has the same scalars and graph, together with the following custom scalar:
This custom scalar chart is simply a layout which combines the original two scalar charts.
Unfortunately the resulting graph is hard to read because both values have the same color. (They are distinguished only by marker.) This is however consistent with TensorBoard's convention of having one color per log.
Explanation
The idea is as follows. You have some group of variables which you want to plot inside a single chart. As a prerequisite, TensorBoard should be plotting each variable individually under the "SCALARS" heading. (This is accomplished by creating a scalar summary for each variable, and then writing those summaries to the log. Nothing new here.)
To plot multiple variables in the same chart, we tell TensorBoard which of these summaries to group together. The specified summaries are then combined into a single chart under the "CUSTOM SCALARS" heading. We accomplish this by writing a "Layout" once at the beginning of the log. Once TensorBoard receives the layout, it automatically produces a combined chart under "CUSTOM SCALARS" as the ordinary "SCALARS" are updated.
Assuming that your "original model" is already sending your variables (as scalar summaries) to TensorBoard, the only modification necessary is to inject the layout before your main iteration loop starts. Each custom scalar chart selects which summaries to plot by means of a regular expression. Thus for each group of variables to be plotted together, it can be useful to place the variables' respective summaries into a separate name scope. (That way your regex can simply select all summaries under that name scope.)
Important Note: The op which generates the summary of a variable is distinct from the variable itself. For example, if I have a variable ns1/my_var, I can create a summary ns2/summary_op_for_myvar. The custom scalars chart layout cares only about the summary op, not the name or scope of the original variable.
Here is an example, creating two tf.summary.FileWriters which share the same root directory. Creating a tf.summary.scalar shared by the two tf.summary.FileWriters. At every time step, get the summary and update each tf.summary.FileWriter.
import os
import tqdm
import tensorflow as tf
def tb_test():
sess = tf.Session()
x = tf.placeholder(dtype=tf.float32)
summary = tf.summary.scalar('Values', x)
merged = tf.summary.merge_all()
sess.run(tf.global_variables_initializer())
writer_1 = tf.summary.FileWriter(os.path.join('tb_summary', 'train'))
writer_2 = tf.summary.FileWriter(os.path.join('tb_summary', 'eval'))
for i in tqdm.tqdm(range(200)):
# train
summary_1 = sess.run(merged, feed_dict={x: i-10})
writer_1.add_summary(summary_1, i)
# eval
summary_2 = sess.run(merged, feed_dict={x: i+10})
writer_2.add_summary(summary_2, i)
writer_1.close()
writer_2.close()
if __name__ == '__main__':
tb_test()
Here is the result:
The orange line shows the result of the evaluation stage, and correspondingly, the blue line illustrates the data of the training stage.
Also, there is a very useful post by TF team to which you can refer.
For completeness, since tensorboard 1.5.0 this is now possible.
You can use the custom scalars plugin. For this, you need to first make tensorboard layout configuration and write it to the event file. From the tensorboard example:
import tensorflow as tf
from tensorboard import summary
from tensorboard.plugins.custom_scalar import layout_pb2
# The layout has to be specified and written only once, not at every step
layout_summary = summary.custom_scalar_pb(layout_pb2.Layout(
category=[
layout_pb2.Category(
title='losses',
chart=[
layout_pb2.Chart(
title='losses',
multiline=layout_pb2.MultilineChartContent(
tag=[r'loss.*'],
)),
layout_pb2.Chart(
title='baz',
margin=layout_pb2.MarginChartContent(
series=[
layout_pb2.MarginChartContent.Series(
value='loss/baz/scalar_summary',
lower='baz_lower/baz/scalar_summary',
upper='baz_upper/baz/scalar_summary'),
],
)),
]),
layout_pb2.Category(
title='trig functions',
chart=[
layout_pb2.Chart(
title='wave trig functions',
multiline=layout_pb2.MultilineChartContent(
tag=[r'trigFunctions/cosine', r'trigFunctions/sine'],
)),
# The range of tangent is different. Let's give it its own chart.
layout_pb2.Chart(
title='tan',
multiline=layout_pb2.MultilineChartContent(
tag=[r'trigFunctions/tangent'],
)),
],
# This category we care less about. Let's make it initially closed.
closed=True),
]))
writer = tf.summary.FileWriter(".")
writer.add_summary(layout_summary)
# ...
# Add any summary data you want to the file
# ...
writer.close()
A Category is group of Charts. Each Chart corresponds to a single plot which displays several scalars together. The Chart can plot simple scalars (MultilineChartContent) or filled areas (MarginChartContent, e.g. when you want to plot the deviation of some value). The tag member of MultilineChartContent must be a list of regex-es which match the tags of the scalars that you want to group in the Chart. For more details check the proto definitions of the objects in https://github.com/tensorflow/tensorboard/blob/master/tensorboard/plugins/custom_scalar/layout.proto. Note that if you have several FileWriters writing to the same directory, you need to write the layout in only one of the files. Writing it to a separate file also works.
To view the data in TensorBoard, you need to open the Custom Scalars tab. Here is an example image of what to expect https://user-images.githubusercontent.com/4221553/32865784-840edf52-ca19-11e7-88bc-1806b1243e0d.png
The solution in PyTorch 1.5 with the approach of two writers:
import os
from torch.utils.tensorboard import SummaryWriter
LOG_DIR = "experiment_dir"
train_writer = SummaryWriter(os.path.join(LOG_DIR, "train"))
val_writer = SummaryWriter(os.path.join(LOG_DIR, "val"))
# while in the training loop
for k, v in train_losses.items()
train_writer.add_scalar(k, v, global_step)
# in the validation loop
for k, v in val_losses.items()
val_writer.add_scalar(k, v, global_step)
# at the end
train_writer.close()
val_writer.close()
Keys in the train_losses dict have to match those in the val_losses to be grouped on the same graph.
Tensorboard is really nice tool but by its declarative nature can make it difficult to get it to do exactly what you want.
I recommend you checkout Losswise (https://losswise.com) for plotting and keeping track of loss functions as an alternative to Tensorboard. With Losswise you specify exactly what should be graphed together:
import losswise
losswise.set_api_key("project api key")
session = losswise.Session(tag='my_special_lstm', max_iter=10)
loss_graph = session.graph('loss', kind='min')
# train an iteration of your model...
loss_graph.append(x, {'train_loss': train_loss, 'validation_loss': validation_loss})
# keep training model...
session.done()
And then you get something that looks like:
Notice how the data is fed to a particular graph explicitly via the loss_graph.append call, the data for which then appears in your project's dashboard.
In addition, for the above example Losswise would automatically generate a table with columns for min(training_loss) and min(validation_loss) so you can easily compare summary statistics across your experiments. Very useful for comparing results across a large number of experiments.
Please let me contribute with some code sample in the answer given by #Lifu Huang. First download the loger.py from here and then:
from logger import Logger
def train_model(parameters...):
N_EPOCHS = 15
# Set the logger
train_logger = Logger('./summaries/train_logs')
test_logger = Logger('./summaries/test_logs')
for epoch in range(N_EPOCHS):
# Code to get train_loss and test_loss
# ============ TensorBoard logging ============#
# Log the scalar values
train_info = {
'loss': train_loss,
}
test_info = {
'loss': test_loss,
}
for tag, value in train_info.items():
train_logger.scalar_summary(tag, value, step=epoch)
for tag, value in test_info.items():
test_logger.scalar_summary(tag, value, step=epoch)
Finally you run tensorboard --logdir=summaries/ --port=6006and you get: