jimduncan/deep_prediction

Accompanying code for 'Deep Prediction of Investor Interest: a Supervised Clustering Approach'.

Deep Prediction of Investor Interest: a Supervised Clustering Approach

This repo contains the accompagnying code for the paper https://arxiv.org/abs/1909.05289. Contribution is two-fold:

1. We first show the usefulness of our approach on simulated data exhibiting the characteristics awaited from our model, and compare it to a LGBM benchmark.
2. We use our approach on an open-source dataset (https://zenodo.org/record/2573031#.XaSQ9_fRY5l) to prove its usefulness on real-world data.

Deep learning code is written in TensorFlow 2.0.

Using this repo

1. Simulation

Run sample_data.py to sample a dataset - the seed is set so as to get the same results as presented in the paper. exnet_run.py, embedmlp_run.py and lgbm_run.py respectively run the ExNet, the EmbedMLP and the LightGBM algorithms on the previously simulated data.

2. IBEX

Run create_data.py to create the dataset described in the article. Run exnet_run.py and lgbm_run.py to respectively run the ExNet and LGBM algorithms on the previously created data. The weights file corresponding to the best ExNet found is included, and can be investiguated using the exnet_analysis.py file.

Requirements

• python==3.7.3
• pandas==0.25.2
• tensorflow==2.0.0
• tensorflow_addons==0.6.0
• numpy==1.17.2
• lightgbm==2.1.2
• ta (https://github.com/bukosabino/ta)
• umap-learn==0.3.10