I decided to build a repository of news headlines: I loaded all ‘New York Times’ headlines since the year 2000 and all Business related News from the ‘Guardian’ into a Sorl Search engine. More details can be found in my prior blog. It has never been the intention to process all documents in one run but the goal was to search for the relevant articles with the help of the search engine and then process Read more…
Initially I used the Solr standard settings but with a big amout of data, I was running out of heap space. Fortunatly it is possible to define the heap space with the SOLR_HEAP environment variable. If you also don’t want to risk to loose your data you should also map the volume /opt/solr/server/solr/mycores Here is the docker-compose.yml file that I am using: version: ‘3.1’ services: solr: image: “solr:alpine” ports: – “8983:8983” volumes: – /srv/solr:/opt/solr/server/solr/mycores environment: Read more…
Today I spent some time to figure out how to use Vegas (which is a plotting library for Scala) in Jupyter with the BeakerX Scala kernel. Here is the result!
Recently I have spent some time to investigate the options to access the history of news articles via an API. I was mainly interested in APIs which can be accessed free of charge. Here is the list of the most useful providers: Guardian – Easy API – Acceptable Rate Limits – Access to over 1,900,000 pieces of content – Free for non-commercial usage New York Times – Provides API to search and separate API to Read more…
I am planning to use the DL4J Doc2Vec implementation for a sentiment analysis. However, I don’t want to start with an empty network but the staring point should be a pre-trained network: The initial trining should be done with the Sentiment140 dataset which can be found at https://www.kaggle.com/kazanova/sentiment140. It contains 1,600,000 tweets extracted using the twitter api. In this Gist I describe how to train and save a DL4J Doc2Vec. The serialized model is available Read more…
The basic goal of a ‘sentiment analysis’ is to classify a given text into positive, negative or neutral. SentiWordNet is a lexical resource for opinion mining. It assigns sentiments to each synset of WordNet which makes it possible to “calculate” an overall sentiment for a text. A SentiWordNet implementation can be found in DL4J in the deeplearning4j-nlp-uima artifact. This demo has been implemented in Scala using a Jupyter BeakerX Notebook.
In this demo I will show how to predict if the closing price of Apple, General Electric and Samsung Electronics is moving up or down. We do this with the help of a Random Forest Classifier. I tried to replicate the result from a research paper authored by Luckyson Khaidem, Snehanshu Saha, Sudeepa Roy Dey which is describing the following concept: Data Preprocessing exponentially smoothing Features Relative Strength Index Stochastic Oscillator Williams %R Moving Average Convergence Read more…
In this blog we show how to forecast the closing price of AAPL using a LSTM RRN network. We use the open, closing, high and low rates and the volume of the current day as input in order to predict the subsequent closing price. This demo has been implemented in Scala using Jupyter with the BeakerX kernel using the following libraries – Investor – DL4j The details can be found in the following Gist.
In recent years most banks are pushing their customers to invest in Funds with the argument that this is more profitable than any other option. As always it is valid to question this statement. As a compelling argument an index chart is presented which shows indeed that the stock market was growing! I tried to run some simulations using the data form my Investor framework to validate this statement: Here is the link to my Gist Read more…
A Long short-term memory (LSTM) network is a special type of a recurrent neural network (RNN). It remembers values over arbitrary time intervals which makes it suited to be used for forecasting of time series. I was struggling with the DL4J implementation of the LSTM to be used to forecast stock data and I was not sure where the issue was: was the data not suited for forecasting ?. did I have a bug in my Iterator which was Read more…