Is ROOT TMVA supporting Recurrent/Convolution Neural Network?

wellatiger · August 6, 2019, 12:33pm

Hello all,

Is ROOT TMVA supporting Recurrent/LSTM/Convolution Neural Network?

Thank you,

bellenot · August 6, 2019, 3:09pm

I think @moneta or @sitong should know …

moneta · August 6, 2019, 5:28pm

Hi

We have currently a vanilla implementation for RNN, but not yet working on GPU, and we are including now LSTM and GRU. They will be available in the master soon, we have already a PR we are reviewing, (see https://github.com/root-project/root/pull/3924)

Convolutional networks are already supported both on CPU and GPU.
An example for convolution rural network is this notebook:

github.com

lmoneta/tmva-tutorial/blob/master/notebooks/TMVA_CNN_Classification.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"tmva_logo.gif\" height=\"20%\" width=\"20%\">\n",
    "\n",
    "# TMVA Classification Example Using a Convolutional Neural Network"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Declare Factory\n",
    "\n",
    "Create the Factory class. Later you can choose the methods\n",
    "whose performance you'd like to investigate. \n",
    "\n",

This file has been truncated. show original

Lorenzo

wellatiger · August 7, 2019, 2:48pm

Thank you very much for your reply,

I just curious how to you prepare the data for the RNN, it should be 3D tensor (batch_size, time_steps, features) and TMVA data-loader accepts TTrees.

Thank you in advance,
Waleed