From 220db379dac97b911188647738ce1a2799faab86 Mon Sep 17 00:00:00 2001
From: Luigi Selmi <luigiselmi@gmail.com>
Date: Wed, 6 Sep 2023 18:18:52 +0200
Subject: [PATCH] minor change

---
 python/iia/crop_production.ipynb | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/python/iia/crop_production.ipynb b/python/iia/crop_production.ipynb
index 09cb20e..d3d3880 100755
--- a/python/iia/crop_production.ipynb
+++ b/python/iia/crop_production.ipynb
@@ -235,10 +235,10 @@
     "\n",
     "* Model: neural network with one hidden layer using the tanh activation function (see [scikit-learn Multi-layer Perceptron](https://scikit-learn.org/stable/modules/neural_networks_supervised.html))\n",
     "* Cost function: quadratic loss (or mean squared error) with momentum regularization\n",
+    "* Training procedure: [leave-one-out](https://scikit-learn.org/stable/modules/cross_validation.html#leave-one-out-loo) cross validation (one example used for test, a subset (10% random examples) used for validation and the rest for training) with [early stopping](https://scikit-learn.org/stable/auto_examples/linear_model/plot_sgd_early_stopping.html)\n",
     "* Optimization algorithm: stochastic gradient descent (fixed learning rate) or Quasi-Newton BFGS (see scikit-learn Multi-layer Perceptron [algorithms](https://scikit-learn.org/stable/modules/neural_networks_supervised.html#algorithms))  \n",
     "* Data pre-processing: mean normalization \n",
-    "* Training procedure: [leave-one-out](https://scikit-learn.org/stable/modules/cross_validation.html#leave-one-out-loo) cross validation (one example used for test, a subset (10% random examples) used for validation and the rest for training) with [early stopping](https://scikit-learn.org/stable/auto_examples/linear_model/plot_sgd_early_stopping.html)\n",
-    "* [Ensemble method](https://scikit-learn.org/stable/modules/ensemble.html): neural network model with different weights initialization and validation set \n",
+    "* [Ensemble method](https://scikit-learn.org/stable/modules/ensemble.html): neural network models with different weights initialization and validation set (bootstrap replicates: sampling with replacement from the training dataset after the test set has been selected)\n",
     "\n",
     "\n",
     "We create an MLP with one hidden layer to approximate a function that represents the yeld of a crop (maize or millet) depending on the yearly values of precipitation, temperature, number of hours with temperatures above 30°C, amount of nitrogen fertilizer, and amount of manure fertilizer. We use a 4-5-1 architecture. We can compute the preactivation of the hidden layer by a matrix multiplications between the units of the input layer and the units of the hidden layer.\n",
@@ -274,7 +274,7 @@
    "metadata": {},
    "source": [
     "## Ensemble method\n",
-    "The methodology can be described algorithmically. The same procedure is used for the two ensembles, one for maize model (4 input units, 5 hidden, 1 output) and one for millet model (5 input units, 5 hidden, 1 output)\n"
+    "The methodology can be described algorithmically. The same procedure is used for the two ensembles, one for maize model (4 input units, 5 hidden, 1 output) and one for millet model (5 input units, 5 hidden, 1 output). The construction of the predictors from the bootstrap samples can be done in parallel on a multicore CPU.\n"
    ]
   },
   {