TensorFlow Question Answers
1 Q-: Explain Tensors?
These are arrays- higher dimension arrays used in computer programming. By using tensors multitude of data is represented in form of numbers. As there are many other libraries i.e. array libraries on the internet like numpy but tensorflow is different as in tensorflow methods are defined to create tensor functions and compute derivatives automatically.
2 Q-: What are tensorflow servables and what do you mean by sources in tensorflow?
Tensorflow servables are central basic units in tensorflow serving. Servables are the objects that are used by clients to perform the computation. The size is flexible and along with this the single servable can include anything from- lookup table- single model to the tuple of inference models.
Sources are responsible for finding and providing the servables known as modules. Zero or more servable streams provided by each source. For loading, one loader is supplied to each servable version.
3 Q-: What are the advantages of TensorFlow over other libraries? What products built in TensorFlow?
Advantages to use tensorflow is as follows-:
Visualization Of Data, Debugging facility, Pipelining, Scalability and many more.
– Tensorflow builds products as-:
- Teachable machine
- Giorgio Camthat
4 Q-: List the advantages and limitations of tensorflow?
Tensorflow has flexible platform.
– It is quite compatible as easily work on CPU and GPU for distributed computing.
– It has auto differentiation- capabilities.
– It includes advanced uphold or support for threads, asynchronous computation and queues.
– It is open source as well as customizable.
– Confliction between theano and GPU memory if import is in the same scope.
– It has no support for open computing language.
– For advance calculus as well as linear algebra prior knowledge is required alongwith better understanding for machine learning.
5 Q-: Name different dashboards in TensorFlow?
6 Q-: Tell the main feature of TensorFlow?
7Q-: Explain different types of Tensors?
There are three types of tensors-:
- Constant Tensor- These are used as constants. In this a node is created that takes a value and doesn’t change it. Constant is created using “tf.constant”.
- Variable Tensor-: Variable tensors – the nodes that provide the current value as output. It means that variable retain the value over the multiple execution of graph.
- Place Holder Tensor-: These tensors are more important than variables. These are responsible for assigning the data and in this the value of node is fetched at the time of execution. As suppose you have some inputs which are depend on some external data. So, if you don’t want to let your graph depend on some real values while developing then the placeholders are very useful kind of datatype. By this you can also build a graph without the data.
It includes five arguments-:
Tf.constant (value, dtype=none, shape= none, name= ‘const’, verify_shape=False)
So the placeholders don’t need any value or initial value, it only needs the datatype as well as shape. So with these things only graph knows how to compute even it doesn’t have any type of values.
8Q-: Explain some ways to load the data into tensorflow?
As Before executing the machine learning algorithm, first step is loading the data into tensorflow.
So, two ways to load the data-:
- Loading of data into memory
- Tensor flow data pipelining
It is convenient method, as all the data is loaded into memory with a single array.
In this there are built-in APIs that helps to load the data and performing the operations and access the machine learning algorithm easily. This method is used in case of large datasets.
9Q-: What are the steps for tensorflow algorithms?
- Importing and generating of data and making a data-pipeline through the placeholders.
- Access or provide the data information through computational graph.
- Evaluate or operate the loss function output.
- Modification of variable by using back propagation.
- Repeat the steps until stopping condition.
10Q-: Tell the name of methods to overcome the situation of over fitting in tensorflow? Explain Tensorflow managers?
Methods to deal with overfitting-:
– Dropout Technique
– Batch Normalization
- TensorFlow managers are responsible for lookup, loading, unloading and lifetime management of servable objects. Tensorflow managers manage or control the lifecycle of servables including-:
– Serving Servables
– Loading Servables
– Unloading Servables
As it is type of abstract class. The syntax is-:
#include < manager.h>
11Q-: Explain Tensorflow servables and serving?
Objects used by clients to perform the computations are known to be as servables which has flexible size. Servables are basic concept of tensorflow serving and single servable contain from lookup table to single model to – tuple of inference models.
Tensorflow serving has also flexible size which is used for the designing of production environments. It has high performance serving system used for machine learning models and with this serving can easily deploys the new algorithms and experiments b y maintaining the same architecture and APIs. Serving provide integration (out-of-the-box) with tensorflow models. To serve different or other types of models and data, it is easily extended whenever required.
12Q-: Tell the Use cases of Tensor Flow?
TensorFlow contains essential tools, it has five cases-:
– Voice/Sound Recognition
– Image Recognition
– Text-Based Applications
– Time Series
– Video Detection
Tensorflow run on different platforms as-: Operating systems such as Windows, OS, Linux, and Cloud web service, mobile OS like IOS and Android.
13Q-: How Tensorflow is better than other libraries?
- Scalability- It provides easily scalable applications (machine learning) and the infrastructure.
- Debugging Facility- Tensorflow specialized debugger is as tfdbg, which lets you view the internal structure of running tensorflow graph as well as states during training and interference.
- Visualization of data-: Visualization tool used is TensorBoard which is used to visualize Tensorflow graphs. As visualizing the graph is very straight-forward in TensorFlow.
- Pipelining- Dataset module of tensorflow i.e. ‘tf.data’ is used to build the well structured pipelines for the text and images.
14Q-: What are tensorflow abstractions?
Abstraction in TensorFlow such as TF-Slim and kereas, actually helps to streamline or efficient construction of data flow graphs which provides the quite high-level access to TensorFlow.
It helps for clean code as well as reduce the length or size of code drastically which results in significantly reduces the development time.
15Q-: What APIs used inside tensorflow project?
APIs inside TensorFlow are python language based, having low-level options for the users like tf.manual or tf.nn.relu helps in build Neural Network Architecture. APIs helps in designing the deep neural network with high level of abstraction.
16Q-: List the API used outside the tensorflow project?
– TensorLayer- It is tensorflow based library with deep learning and reinforcement. Tensorlayer is designed for researchers and engineers that provide huge collection of customizable neural layers/functions that are critical for building real-world AI applications.
– TFLearn- It provides high level API that makes neural network building. It is fully compatible with tensorflow and it is indicated as tf.contrb.learn.
– Sonnet- It is library that builts on top of TensorFlow which creates complex neural network. Sonnet is part of Google’s DeepMind project that feature a modular approach.
– Pretty Tensor- It delivers high-level builder API, and offers thin wrappers that helps to build multi-layer neural networks easily.
17Q-: What is difference between tf.variable and tf.placeholder?
TF.VARIABLE defines the values that are changed or modified with time whereas TF.PLACEHOLDER defines the input data which doesn’t change with time.
– TF.VARIABLE requires the initial value at definition time whereas TF.PLACEHOLDER doesn’t require initial value at the time of definition .
18Q-: Explain Distribution dashboard in tensorflow?
It is the way to visualize the graph (histogram) data from ‘tf.summary.histogram’. Distribution dashboard displays the high level statistics with this each line on the chart represents percentile in distribution over the data. Such as- bottom line tells that how minimum value varies over time and middle line shows the variation of median. Reading is actually done from top to bottom i.e. maximum percentage to minimum percentage.
19Q-: Explain the Image dashboard in tensorboard?
It shows the latest image in case of every tag, it displays png files that are saved via tf.summary.image. Dashboard is configured in a way in which each row belongs to different tag and column corresponds to run. It also supports arbitrary pngs that can be used to submerge custom visualizations for example “matplotlib scatterplots” into the tensorboard.
20Q-: Explain the Graph Explorer in TensorFlow?
For visulazation of TensorBoard graph the Graph Explorer is used which does the inspection for tensorflow model. If you want to use graph visualize efficiently then you should use name scopes for the grouping of ops in graph (hierarchically) or otherwise the graph will be challenged to decipher.
21Q-: List the different dashboards in Tensorflow?
There are different dashboards to perform various tasks as-:
- Histogram Dashboard
- Scalar Dashboard
- Image Dashboard
- Graph Dashboard
- Distributor Dashboard
- Audio Dashboard
- Text Dashboard
22Q-: Difference between TensorFlow and PyTorch?
|It is based on Theano Library||It is software-based on Torch library.|
|It is developed by google.||It is developed by facebook.|
|It doesn’t have any option at run time.||It contains the computational graph at runtime.|
|It follows the Tensorboard to visualize the machine learning model.||It doesn’t contain any visualization feature.|
23Q-: What are tensorflow decision forests?
Decision forests are a collection of algorithms (state-of-the-art) for serving, training as well as interpretation of decision forest models.
– Library contains- supports classification, regression, keras models, ranking.
- Loading the dataset in dataframe as-:
- Converting dataset into tensorflow dataset as-:
- Training the model is as-:
- Look the whole model-:
- Evaluating model as-:
Train_df = pd.read_csv(“project/train.csv”)
Test_df = pd.read_csv(“project/test.csv”)
Train_ds = tfdf.keras.pd_dataframe_to_tf_dataset(train_df, label = “my_label”)
Test_ds = tfdf.keras.pd_dataframe_to_tf_dataset(test_df, label = “my_label”)
24Q-: Explain tensorflow ranking?
It is used to learn ranking techniques on tensorflow platform. It contains following components-:
– Loss functions including pairwise, listwise, pointwise losses.
– Ranking metrics used like Normalized Discounted Cumulative Gain (NDCG) and Mean Reciprocal Rank (MRR)
– GroupWise scoring functions also known as multi-item functions.
– Lambdaloss implementation which is used for direct ranking metric optimization.
– It supports unbiased learning to rank.
25Q-: Difference between CNN and RNN?
|It is known as convolution neural network.||It is recurrent neural network.|
|It is known as – feed forward model.||It is used for the series of data|
|It is a memory less model.||It requires memory to store previous inputs.|
|Sequential data is not handled by CNN.||It can handle the sequential data.|
|It is used for image recognition.||It is used for text recognition.|
|It includes the handling of fixed length of input and output.||It handles the arbitrary lengths of input/output.|
|It has more feature compatibility.||It has less feature compatibility.|
|It handles the permanent data.||It handles the temporary data.|