Modular Neural Network Protyping for Stock Market Prediction
Project description
MLProto: Modular Prototyping Tool for LSTM Machine Learning Models
Usage
Video overview coming
The ProtoMake Script
The ProtoMake script combines the Proto and Data modules into one to create an easy, convenient, and modular neural network prototyping tool for LSTM machine learning models. The script will take the user's desired parameters and create, train, and evaluate a model fitting said parameters. This allows the user to quickly analyze model prototypes, make adjustments, and iterate on model designs.
Arguments:
Positional:
key ---- User's Alpha_Vantage API key
identifier ---- Ticker symbols to create models for
target ---- Column number of target values
Optional:
-depth ---- number of layers to include in the neural network (def: 1)
-node_counts ---- list of node counts for layers (len(node_counts) must equal depth)
-batch ---- batch size of input data set (def: [100])
-test_size ---- proportion of dataset to use as validation (def: .2)
-loss ---- identifier string of keras-supported loss function to be used in training (def: mse)
-learning_rate ---- learning rate to be used by the Adam optimizer
-epochs ---- maximum number of epochs to train the model (def: 100)
-model_in ---- file path of pre-made model to load
--early_stop ---- flag deciding whether to apply early stopping (patience 5) to the training phase
--plots ---- flag deciding whether to save loss, input, and prediction graphs
--normalize ---- flag deciding whether or not to normalize input data
Usage Example:
ProtoMake test.csv test_model --early_stop --plots
The above command will create, train, and evaluate a model for the data in test.csv. It saves a model test_model.h5 in directory ./models/ and input, loss, and prediction graphs in the directory ./plots/ for analysis.
The Stocker Module
If you would like to use your own data pipelines as inputs, the Stocker and data helper modules can be used separately from the main script.
from MLProto import *
""" Data operations (assign data, pred_data)
______________________________________________
"""
stkr = Proto(data)
stkr.train(25, True, True))
stkr.evaluate()
stkr.predict_data(pred_data)
The above code will take prepared data, create a stocker instance for the FORD ticker, train for 25 epochs, save the model to the models folder as FORD.h5 and predict a data point based on the user's prepared prediction data.
The Data module
This module includes the data operation helper functions used by Stocker.
single_step_data takes a full dataset and creates a single-step timeseries dataset from it for input into an LSTM model.
Contributions
Please send pull requests! I am a full-time student, so development and support for Stocker will likely be slow with me working alone. I welcome any and all efforts to contribute!
License
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.