scripts/builtin/sherlockPredict.dml

#-------------------------------------------------------------
#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
#
#-------------------------------------------------------------

# This function implements prediction and evaluation phase of Sherlock:
# Split feature matrix into four different feature categories and predicting the class probability
# on the respective features. Then combine all predictions for final predicted probabilities.
# A Deep Learning Approach to Semantic Data Type Detection.
# [Hulsebos, Madelon, et al. "Sherlock: A deep learning approach to semantic data type detection."
# Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining.
# 2019.]
#
# INPUT:
# -------------------------------------------------------------------------------------------
# X     matrix of values which are to be classified
# cW    weights (parameters) matrices for character distribtions
# cb    biases vectors for character distribtions
# wW    weights (parameters) matrices for word embeddings
# wb    biases vectors for word embeddings
# pW    weights (parameters) matrices for paragraph vectors
# pb    biases vectors for paragraph vectors
# sW    weights (parameters) matrices for global statistics
# sb    biases vectors for global statistics
# fW    weights (parameters) matrices for  combining all trained features (final)
# fb    biases vectors for combining all trained features (final)
# -------------------------------------------------------------------------------------------
#
# OUTPUT:
# ------------------------------------------------------------------------------------------
# probs  class probabilities of shape (N, K)
# ------------------------------------------------------------------------------------------

source("scripts/nn/examples/sherlockNet.dml") as sherlockNet

m_sherlockPredict = function(Matrix[Double] X,
            Matrix[Double] cW1, Matrix[Double] cb1,
            Matrix[Double] cW2, Matrix[Double] cb2,
            Matrix[Double] cW3, Matrix[Double] cb3,
            Matrix[Double] wW1, Matrix[Double] wb1,
            Matrix[Double] wW2, Matrix[Double] wb2,
            Matrix[Double] wW3, Matrix[Double] wb3,
            Matrix[Double] pW1, Matrix[Double] pb1,
            Matrix[Double] pW2, Matrix[Double] pb2,
            Matrix[Double] pW3, Matrix[Double] pb3,
            Matrix[Double] sW1, Matrix[Double] sb1,
            Matrix[Double] sW2, Matrix[Double] sb2,
            Matrix[Double] sW3, Matrix[Double] sb3,
            Matrix[Double] fW1, Matrix[Double] fb1,
            Matrix[Double] fW2, Matrix[Double] fb2,
            Matrix[Double] fW3, Matrix[Double] fb3)
      return (Matrix[Double] probs) {

  rows = nrow(X)

  cprobs = sherlockNet::predict(X[1:rows, 224:1183], cW1, cb1, cW2, cb2, cW3, cb3)
  wprobs = sherlockNet::predict(cbind(X[1:rows, 13:212], X[1:rows, 1188]), wW1,wb1, wW2, wb2, wW3, wb3)
  pprobs = sherlockNet::predict(X[1:rows, 1189:1588], pW1, pb1, pW2, pb2, pW3, pb3)
  sprobs = sherlockNet::predict(cbind(X[1:rows, 1:12], X[1:rows, 213:223], X[1:rows, 1184:1187]), sW1, sb1, sW2, sb2, sW3, sb3)

  first_predictions = cbind(cprobs, wprobs, pprobs, sprobs)
  #final training
  probs = sherlockNet::predict(first_predictions, fW1, fb1, fW2, fb2, fW3, fb3)
}

# Evaluates the performance of the network.
# ---------------------------------------------------------------------------------------------
# NAME         TYPE      DEFAULT  MEANING
# ---------------------------------------------------------------------------------------------
# probs        Matrix             class probabilities of shape (N, K) (one-hot encoded)
# Y            Matrix             target matrix of shape (N, K)
# ---------------------------------------------------------------------------------------------
# loss         double             scalar loss, of shape (1)
# accuracy     double             scalar accuracy, of shape (1)
# f1 score     double             scalar f1 score, of shape (1)
# precision    double             scalar precission, of shape (1)
# recall       double             scalar recall, of shape (1)

eval = function(Matrix[Double] probs, Matrix[Double] Y)
  return (double loss, double accuracy, double f1, double precision, double recall) {

  [loss, accuracy, f1, precision, recall] = sherlockNet::eval(probs, Y)
}