randallsalvares Goto Github PK

blackfriday-sales-prediction

breast-cancer-detection-

Find Malignant and benign cells

credit-card-fraud-detection

find fraudsters

deep-learning-v2-pytorch

Projects and exercises for the latest Deep Learning ND program https://www.udacity.com/course/deep-learning-nanodegree--nd101

face-detection-challenge

Identify faces and locate the coordinates of the rectangle in the image

facedetect

face_predictor

Basic face detection

predicting-stock-price-machine-learningnd-deep-learning-techniques

rnn-walkthrough

titanic-survivor-predictions

turbo-fan-engine-run-to-failure-predictions-using-pytorch-lstm-networks

The need for deep learning approaches in SMART manufacturing is gaining traction, so the need for highly optimized models is in high demand as deep learning algorithms have shown an increased accuracy in predicting the time-series related dataset problems due to its complex architecture compared to conventional machine learning techniques. In this repo we can see how a improved version of the Recurrent Neural Network (RNN) known as the LSTM or Long Short-Term Memory Network works in producing highly optimized models for prognostic related challenges. I am using LSTM architecture on a PyTorch framework backend to build a model which is tested on four different run to failure datasets provided by the NASA repository on turbofan degradation for aircrafts. The repo starts by highlighting the benefits and developments in today’s world done by deep learning approaches especially in the Manufacturing industry. A brief introduction is also provided about how the data is generated and pre-processed for model training. This is then followed by showing a step by step approach of how the LSTM PyTorch model is designed with various parameters and clear results are displayed in tabular forms of the test results for 3 different optimizers, which are RMSprop, SGD and ADAM. The results show after comparing the RMSE values for each optimizer trained models that the RMSprop is clearly better. This is followed by a further hyper-parameter optimization to improve the current RMSE results also displayed in tables below. The repository ends on showing how Transfer Learning can be utilised to further deep learning approaches by using already trained models to predict newer datasets.

value_predictor.py

takes a list of an even number of value and is of at minimum length of 6 sequential and increasing elements and find the most likely middle value.