TensorFlowPlayGround

Tensorflow codes written as part of Advanced Machine Learning Course Work

SimpleTensors.py

This code illustrates the use the symbolic constants and variables in tensorflow, interactive sessions which feed the data to the variables at runtime using two different approaches

LinearRegressorEstimator

1. LinearRegressorEstimator.py
2. dataset.csv

A Linear Regressor is trained with a linear dataset using Tensorflow. This code is mainly aimed to illustrates the use of:

• Dataset API
  • Provides an efficient input pipeline
  • Involves creating dataset instance from the data, creating an iterator and consuming data
• Pre-made Estimator
  • Provide a much higher conceptual level than the base TensorFlow APIs
  • Abstracts the creation of computational graph or sessions since Estimators handle all

EMNIST_DNNClassifier

1. DNN.py
2. Report.txt

A DNN Classifier which is a Premade Estimator is used to predict the class of EMNIST images.

Dataset - https://www.nist.gov/itl/iad/image-group/emnist-dataset

This code illustrates the following :

• Reading data directly from the ubyte file
• Data Preprocessing
• Creating an instance of Dataset API
• Using Shuffle and Pre-fetch functinalities of TensorFlow
• Using Pre-Made DNN Estimator to train and test
• Reporting metrics per class
• Reporting overall metrics

Referred to http://cjalmeida.net/post/tensorflow-mnist/

Selfi-Dataset

DataSet - http://crcv.ucf.edu/data/Selfie/

• Selfie dataset contains 46,836 selfie images annotated with 36 different attributes divided into several categories as follows.
• Gender: is female. Age: baby, child, teenager, youth, middle age, senior. Race: white, black, asian. Face shape: oval, round, heart. Facial gestures: smiling, frowning, mouth open, tongue out, duck face. Hair color: black, blond, brown, red. Hair shape: curly, straight, braid. Accessories: glasses, sunglasses, lipstick, hat, earphone. Misc.: showing cellphone, using mirror, having braces, partial face. Lighting condition: harsh, dim.

1. multiclass.ipynb

• This file contains the code to:
  • analysis of datapoints and augment images with lower distribution
  • Resnet50 architecture used as a base and last few layers added to it and trained
  • made to predict 36 attributes

2. multitask.ipynb

• This file has two parts:
  • 2 output heads
    • input images and predict popularity score along with 7 attributes
  • 2 input heads
    • input images and popularity scores, predict 36 attributes

3. popularity_score.ipynb

• This file constains predicting the class of an image (great,avg,poor - based on popularity score- using simple fraction ethod to group classes.)

4. popularity_score_qcut.ipynb

• This file contains predicting the class of an image (great,avg,poor - based on popularity score- using q-cut approach to group classes.)

5. analysis.ipynb

• This file contains the code to predict the importance of a particular features