Software Project - AI Prediction Tests



  • Software Project - AI Prediction Tests


    In order to get access to our GitHub repository, a test is mandatory to ensure that your skills are sufficient to make contributions. As our code is the result of years of AI research we have to restrict access to proven experts only.


    Test 1

    Build a Sine wave predictor which reads 1200 data points of sine waves from an input file and makes a 260 step forward prediction. The input file should have one data point per row. You can choose the amount of data points per oscillation, but ensure that at least a couple of oscillations are available for training. To predict a single step forward is simple, but we aim to build a predictor who can go far into the future without any further training after the initial training. So you need to make 260 forward predictions at once (not point to point) without any training in between.

    Plot the result in a chart (you can use a different tool for plotting).

    Example Chart:

    Sine Predictor.png

     

    • Choose one of these languages: C99, C++ or Python with Tensorflow API 2.0 (not 1.5).
    • For C99 and C++ ensure that you use AVX, not AVX2 or AVX-512
    • Your project needs to be uploaded to GitHub with a reference to this test.


    Scoring will be done by testing datasets with different oscillations, so any “hardcoding” won’t make you pass.

     

    Test 2

    Predict the Apple stock price 260 days into the future.

    Dataset:
    https://finance.yahoo.com/quote/AAPL/history?period1=345427200&period2=1571698800&interval=1d&filter=history&frequency=1d

    • Time Period: Max
    • Show: Historical Prices
    • Frequency: Daily

    Than press Apply and click on “Download Data” to get the result in a file.

    Manually remove invalid rows and remove all columns except Close price.
    So the input dataset will be a text file with rows which each have a single value, like:

    0.513393
    0.486607
    0.450893
    […]

    Remove the last 260 rows of the dataset, but keep the very last price in your program as that is your target to predict and to verify the accuracy of your prediction model.

    Now build a prediction model using LSTM (to run on a CPU, not GPU) using either C99, C++ or Python with Tensorflow 2.0 (not 1.5) to predict a single stock price 260 days into the future. This can be a single value forecast, no need to predict a full trendline to save processing time.

    Plot a simple chart to show the past 1200 training datasets and the predicted value 260 days ahead and the real value to compare for this test (you can use a different tool for plotting).


    Example Chart:

    260.png

    • Choose one of these languages: C99, C++ or Python with Tensorflow API 2.0 (not 1.5).
    • For C99 and C++ ensure that you use AVX, not AVX2 or AVX-512
    • Your project needs to be uploaded to GitHub with a reference to this test.


    Validation will be done by using a dataset of a different well known company to see if your model still performs well.

     

    Next Steps

    • We will score your projects, and once passed:

    • You will present present your solutions in a group call.

    • We will go through our software documentation to make you familiar with our software projects.

    • You get access to our private Github repository and we will share a Dropbox folder with you to get the required databases.

    • Your first project will be similar to Test 2, but one level more complex.

     

    Role Details

    You would join the AI team and do research on new AI models like seq2seq attention networks, and also create and improve our long term AI models. You will work with an SQLite database with 3 million companies and predict the price of each company in one year, as good as possible, based on 5 year past stock price data.

    Please note that all positions we have are currently voluntary based with no compensation.


Log in to reply