The document describes a method for using a convolutional neural network to classify signals from the Kepler telescope as exoplanets or other phenomena. It involves:
1. Dividing the Kepler telescope light curve data into training, validation, and test sets.
2. Developing a CNN model with convolutional layers to analyze global and local light curves, and a fully connected layer to combine the outputs.
3. Training the model by passing hyperparameters like the number of epochs, data loaders, model, loss criterion, and optimizer to optimize the model's predictions against the validation set.
The document describes a method for using a convolutional neural network to classify signals from the Kepler telescope as exoplanets or other phenomena. It involves:
1. Dividing the Kepler telescope light curve data into training, validation, and test sets.
2. Developing a CNN model with convolutional layers to analyze global and local light curves, and a fully connected layer to combine the outputs.
3. Training the model by passing hyperparameters like the number of epochs, data loaders, model, loss criterion, and optimizer to optimize the model's predictions against the validation set.
The document describes a method for using a convolutional neural network to classify signals from the Kepler telescope as exoplanets or other phenomena. It involves:
1. Dividing the Kepler telescope light curve data into training, validation, and test sets.
2. Developing a CNN model with convolutional layers to analyze global and local light curves, and a fully connected layer to combine the outputs.
3. Training the model by passing hyperparameters like the number of epochs, data loaders, model, loss criterion, and optimizer to optimize the model's predictions against the validation set.
Ch.Surya Meghana (16131A1222) D.Mounika (16131A1226) K.Gayathri Sai Prathyusha (16131A1255) ABSTRACT Our moto is to determine the frequency of planets orbiting Sun-like stars, but these planets are on the very edge of the mission's detection sensitivity. Accurately determining the occurrence rate of these planets will require automatically and accurately assessing the likelihood that individual candidates are indeed planets, even at low signal-to-noise ratios. We present a method for classifying potential planet signals using deep learning, a class of machine learning algorithms that have recently become state-of-the-art in a wide variety of tasks. We train a deep convolutional neural network to predict whether a given signal is a transiting exoplanet or a false positive caused by astrophysical or instrumental phenomena. WHAT IS EXOPLANET An exoplanet or extra solar planet is a planet outside the Solar System. Planets that orbit around other stars are called exoplanets. Exoplanets are very hard to see directly with telescopes. They are hidden by the bright glare of the stars they orbit. WHAT IS LIGHT CURVE The light curve is a graph that shows the brightness of the star over time, and is the measurement Kepler makes to discover exoplanets. The dip in light that happens when the planet passes in front of the star is called the "transit." Transits give information about the planet's size and orbit. ABOUT DATASET In this project we will be taking the data set which is produced by the Kepler telescope. The data set is in the form of numpy array which consists of light curve, planet candidate, astrophysical false positive (AFP), and nontransiting phenomenon (NTP). We randomly partitioned our data into three subsets: training (80%), validation (10%), and test (10%). We used the validation set during development to choose model hyper-parameters, and we used the test set to evaluate final model performance. PYTORCH PyTorch is an open source machine learning library based on the Torch library, used for applications such as computer vision and natural language processing. It is primarily developed by Facebook's AI Research lab. It is free and open-source software released under the Modified BSD license. We are using PyTorch in this project development because it is much easier to use than tensor flow. ABOUT THE MODEL In order to detect the exoplanets we are going to develop a convolutional neural network The workflow of the model is explained in further steps.. CNN
A CNN typically consists of convolutional layers and pooling
layers The input to a (one-dimensional) convolutional layer is a stack of K vectors of length , and the output is a stack of L vectors (l = 1,2, ..., L) of length ni. The operation that takes the stack of K input vectors to the lth output vector is called a feature map and is defined by the operation
Where * is the discrete cross-correlation operation
(colloquially called "convolution"), is a vector of length mi of learned parameters called the convolution kernel or filter, is a vector of length ni of learned bias parameters, and is an element wise activation function. STEP1 The first step of the project is we divide the data set into three parts. That is the global light curves, local light curves, the info . This is done by constructing a class and then the obtained output is reshaped to make a single row for better processing. STEP2 In this we are constructing the base architecture for our model. We will be constructing the model with three convolution layers, one for the global light curves, one for local light curves. In the third layer it is a fully connected layer i.e. it is generated by the combination of the outputs generated from the first two layers. STEP3 After the architecture model is prepared , our model is ready to be trained. In order to train our model and get the precise accuracy for the detection we should pass some parameters to our training model. They are: 1: number of epochs for training 2:data loader for training set 3: data loader for validation set 4: model being trained 5: criterion for calculating the loss 6: optimizer used for training WHAT IS EPOCH? Epoch is nothing but the number of times the model should train itself so that it gets better accuracy in the prediction. The higher the epoch value the greater the accuracy.
Criterion for calculating the loss:
It is the amount of the limit that we can able to face the loss.
Optimizer used for training:
In this model we will be using Adam optimizer Adam is an optimization algorithm that can be used instead of the classical stochastic gradient descent procedure to update network weights iterative based in training data. CONCLUSION After all the requirements are gathered now it is time for the execution. During the execution for every epoch iteration the epoch value loss that is its validation loss, epoch value accuracy(validation accuracy) and the average precision is calculated and displayed as output. Basing on the above values generated the final predictions and ground truths for validation set are obtained. OUTPUT OUTPUT OUTPUT