From 5ecec6d53a4f83b7331c3b3f14bf97b9ffa8de2d Mon Sep 17 00:00:00 2001
From: Mikolaj Rybinski <mikolaj.rybinski@id.ethz.ch>
Date: Mon, 9 Sep 2019 08:16:54 +0200
Subject: [PATCH] minor corrections post review by Uwe
---
01_introduction.ipynb | 2 +-
02_classification.ipynb | 2 +-
03_overfitting_and_cross_validation.ipynb | 83 +++++++++++++----------
3 files changed, 48 insertions(+), 39 deletions(-)
diff --git a/01_introduction.ipynb b/01_introduction.ipynb
index b478166..7ffecd1 100644
--- a/01_introduction.ipynb
+++ b/01_introduction.ipynb
@@ -1408,7 +1408,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "##### What happened?\n",
+ "<div style=\"font-weight: bold; font-size: 200%;\">What happened?</div>\n",
"\n",
"Why were not all labels predicted correctly?\n",
"\n",
diff --git a/02_classification.ipynb b/02_classification.ipynb
index 2a2e431..20ec770 100644
--- a/02_classification.ipynb
+++ b/02_classification.ipynb
@@ -1353,7 +1353,7 @@
"Most cases have higher dimensions than 2 or 3 and visual inspection can be difficult. Thus, engineering features as we did in the 2D examples becomes tricky.\n",
"\n",
"<div class=\"alert alert-block alert-warning\"><p><i class=\"fa fa-warning\"></i> \n",
- "<span style=\"font-size: large\">General recommendations for feature engineering</span>\n",
+ "<span style=\"font-weight: bold; font-size: 125%;\">General recommendations for feature engineering</span>\n",
"<ul>\n",
"<li>use descriptive statistics (mean, standard deviations, higher order statistics), as well as histograms if applicable;</li>\n",
"<li>polynomial features (e.g. extend <code>x, y</code> to <code>x, y, x * y, x ** 2, y ** 2</code>) (see below);</li>\n",
diff --git a/03_overfitting_and_cross_validation.ipynb b/03_overfitting_and_cross_validation.ipynb
index 99b495f..7e3d3f6 100644
--- a/03_overfitting_and_cross_validation.ipynb
+++ b/03_overfitting_and_cross_validation.ipynb
@@ -1171,14 +1171,17 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "300\n",
- "0.49666666666666665\n",
+ "# Whole dataset \n",
+ "number of all samples: 300\n",
+ "proportion of yummy samples: 0.49666666666666665\n",
"\n",
- "240\n",
- "0.49583333333333335\n",
+ "# Cross-validation dataset \n",
+ "number of all samples: 240\n",
+ "proportion of yummy samples: 0.49583333333333335\n",
"\n",
- "60\n",
- "0.5\n"
+ "# Validation dataset \n",
+ "number of all samples: 60\n",
+ "proportion of yummy samples: 0.5\n"
]
}
],
@@ -1198,14 +1201,17 @@
"\n",
") = train_test_split(features, labels, test_size=0.2, stratify=labels, random_state=42)\n",
"\n",
- "print(len(labels))\n",
- "print(sum(labels == 1)/len(labels))\n",
+ "print(\"# Whole dataset \")\n",
+ "print(\"number of all samples:\", len(labels))\n",
+ "print(\"proportion of yummy samples:\", sum(labels == 1)/len(labels))\n",
"print()\n",
- "print(len(labels_crosseval))\n",
- "print(sum(labels_crosseval == 1)/len(labels_crosseval))\n",
+ "print(\"# Cross-validation dataset \")\n",
+ "print(\"number of all samples:\", len(labels_crosseval))\n",
+ "print(\"proportion of yummy samples:\", sum(labels_crosseval == 1)/len(labels_crosseval))\n",
"print()\n",
- "print(len(labels_validation))\n",
- "print(sum(labels_validation == 1)/len(labels_validation))"
+ "print(\"# Validation dataset \")\n",
+ "print(\"number of all samples:\", len(labels_validation))\n",
+ "print(\"proportion of yummy samples:\", sum(labels_validation == 1)/len(labels_validation))"
]
},
{
@@ -1225,21 +1231,21 @@
"output_type": "stream",
"text": [
"OPTIMIZE HYPERPARAMETERS\n",
- "score = 0.717 +/- 0.108, C = 0.1, gamma = 0.1\n",
- "score = 0.796 +/- 0.063, C = 0.1, gamma = 1.0\n",
- "score = 0.862 +/- 0.056, C = 0.1, gamma = 10.0\n",
+ "score = 0.728 +/- 0.107, C = 0.1, gamma = 0.1\n",
+ "score = 0.791 +/- 0.090, C = 0.1, gamma = 1.0\n",
+ "score = 0.824 +/- 0.057, C = 0.1, gamma = 10.0\n",
"score = 0.504 +/- 0.008, C = 0.1, gamma = 100.0\n",
- "score = 0.825 +/- 0.048, C = 1.0, gamma = 0.1\n",
- "score = 0.908 +/- 0.036, C = 1.0, gamma = 1.0\n",
- "score = 0.900 +/- 0.046, C = 1.0, gamma = 10.0\n",
- "score = 0.784 +/- 0.074, C = 1.0, gamma = 100.0\n",
- "score = 0.933 +/- 0.050, C = 10.0, gamma = 0.1\n",
- "score = 0.958 +/- 0.037, C = 10.0, gamma = 1.0\n",
- "score = 0.909 +/- 0.040, C = 10.0, gamma = 10.0\n",
- "score = 0.780 +/- 0.072, C = 10.0, gamma = 100.0\n",
+ "score = 0.837 +/- 0.094, C = 1.0, gamma = 0.1\n",
+ "score = 0.875 +/- 0.042, C = 1.0, gamma = 1.0\n",
+ "score = 0.871 +/- 0.054, C = 1.0, gamma = 10.0\n",
+ "score = 0.775 +/- 0.051, C = 1.0, gamma = 100.0\n",
+ "score = 0.908 +/- 0.031, C = 10.0, gamma = 0.1\n",
+ "score = 0.950 +/- 0.031, C = 10.0, gamma = 1.0\n",
+ "score = 0.891 +/- 0.034, C = 10.0, gamma = 10.0\n",
+ "score = 0.780 +/- 0.057, C = 10.0, gamma = 100.0\n",
"\n",
"BEST RESULT CROSS VALIDATION\n",
- "score = 0.958 +/- 0.037, C = 10.0, gamma = 1.0\n"
+ "score = 0.950 +/- 0.031, C = 10.0, gamma = 1.0\n"
]
}
],
@@ -1263,7 +1269,7 @@
" for gamma in SVC_gamma_values:\n",
" classifier = SVC(C=C, gamma=gamma)\n",
" test_scores = cross_val_score(classifier, features_crosseval, labels_crosseval,\n",
- " scoring=\"accuracy\", cv=cross_validator)\n",
+ " scoring=\"accuracy\", cv=cross_validator) # cv arg is now different\n",
" print(\"score = {:.3f} +/- {:.3f}, C = {:5.1f}, gamma = {:5.1f}\".format(\n",
" test_scores.mean(), test_scores.std(), C, gamma))\n",
" results.append((\n",
@@ -1506,25 +1512,27 @@
"output_type": "stream",
"text": [
"OPTIMIZE SETTINGS\n",
- "score = 0.921 +/- 0.039, C = 1.0, penalty = l1\n",
- "score = 0.925 +/- 0.041, C = 10.0, penalty = l1\n",
- "score = 0.925 +/- 0.041, C = 100.0, penalty = l1\n",
- "score = 0.929 +/- 0.037, C = 1000.0, penalty = l1\n",
- "score = 0.825 +/- 0.083, C = 1.0, penalty = l2\n",
- "score = 0.913 +/- 0.039, C = 10.0, penalty = l2\n",
- "score = 0.917 +/- 0.037, C = 100.0, penalty = l2\n",
- "score = 0.925 +/- 0.041, C = 1000.0, penalty = l2\n",
+ "score = 0.883 +/- 0.085, C = 1.0, penalty = l1\n",
+ "score = 0.904 +/- 0.077, C = 10.0, penalty = l1\n",
+ "score = 0.904 +/- 0.077, C = 100.0, penalty = l1\n",
+ "score = 0.904 +/- 0.077, C = 1000.0, penalty = l1\n",
+ "score = 0.804 +/- 0.079, C = 1.0, penalty = l2\n",
+ "score = 0.888 +/- 0.076, C = 10.0, penalty = l2\n",
+ "score = 0.908 +/- 0.074, C = 100.0, penalty = l2\n",
+ "score = 0.904 +/- 0.077, C = 1000.0, penalty = l2\n",
"\n",
"BEST RESULT CROSS VALIDATION\n",
- "score = 0.929 +/- 0.037, C = 1000.0, penalty = l1\n",
+ "score = 0.908 +/- 0.074, C = 100.0, penalty = l2\n",
"\n",
"VALIDATION\n",
- "score = 0.850\n"
+ "score = 0.933\n"
]
}
],
"source": [
"from sklearn.linear_model import LogisticRegression\n",
+ "from sklearn.svm import SVC\n",
+ "from sklearn.model_selection import cross_val_score\n",
"\n",
"beer = pd.read_csv(\"data/beers.csv\")\n",
"beer_eval = pd.read_csv(\"data/beers_eval.csv\")\n",
@@ -1534,13 +1542,14 @@
"labels = all_beer.iloc[:, -1]\n",
"\n",
"(\n",
- "\n",
" features_crosseval, \n",
" features_validation, \n",
" labels_crosseval, \n",
" labels_validation,\n",
"\n",
- ") = train_test_split(features, labels, test_size=0.2, stratify=labels, random_state=42)\n",
+ ") = train_test_split(features, labels, test_size=0.2, stratify=labels, random_state=42) \n",
+ "\n",
+ "# TRY random_state 43 OR 1 INSTEAD OF 42!\n",
"\n",
"cross_validator = StratifiedKFold(n_splits=10, shuffle=True, random_state=42)\n",
"\n",
--
GitLab