Agatha is a tool to help you predict future prices (open, close) or daily volume for any given stock ticker.

|

Probably not.

Agatha uses an LSTM network to predict close prices for a user-specified number of days in the future. The training data is downloaded via Alpha Vantage.

• python 3.5 or higher

There are two ways to install agatha.

The easiest way to install agatha is via pip:

pip install agatha

Note: keep in mind that this requires python 3.5 or higher. Another Note: If you want the latest version build from sources.

Clone this repository. Inside the Agatha folder, create the agatha package using

python setup.py sdist

Then install using pip.

pip install dist/*

If you use anaconda, you can load the conda environment using the environment.yml file in resources/conda and running conda env create -f environment.yml

First, import agatha's functions

from agatha import getOrTrainModel, predictFuture

Then get an API key from Alpha Vantage. To train a model for a particular ticker, use

model = getOrTrainModel(alpha_vantage_api_key, ticker, attribute, alphavantage_data,
						model_data, weights_data, epochs=epochs, look_back=look_back)

where

• ticker is the stock ticker
• attribute is the stock attribute to predict (open, close, volume),
• alphavantage data is downloaded as a csv and then pickled (saved as .pkl)
• the model_data is saved as json
• the weights file is saved as .h5

Predictions for future close prices for a stock can have output type as json or plot (pyplot, as shown in graphs above)

prediction_output = predictFuture(model, num_days_to_predict, ouptut_type)

Example:

model = getOrTrainModel('adsfadsfasdf', 'GE', 'GE.pkl', 'open', 'model.json', 'weights.h5')
prediction_output = predictFuture(model, 2, 'json')

Example output JSON from predictFuture:

{
   "ticker":"GE",
   "column":"open",
   "predictions":[
      {
         "day":"1",
         "price":"8.009521"
      },
      {
         "day":"2",
         "price":"8.117293"
      }
}

Refer to app.py, for a working example.

• Allow other sources of historical data (including cryptocurrencies)
• Any suggestions?