forked from tensorflow/tfjs-examples
-
Notifications
You must be signed in to change notification settings - Fork 1
/
index.js
167 lines (150 loc) · 5.61 KB
/
index.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
/**
* @license
* Copyright 2018 Google LLC. All Rights Reserved.
* Licensed 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.
* =============================================================================
*/
import * as tf from '@tensorflow/tfjs';
import {WebsitePhishingDataset} from './data';
import * as ui from './ui';
import * as utils from './utils';
function falsePositives(yTrue, yPred) {
return tf.tidy(() => {
const one = tf.scalar(1);
const zero = tf.scalar(0);
return tf.logicalAnd(yTrue.equal(zero), yPred.equal(one))
.sum()
.cast('float32');
});
}
function trueNegatives(yTrue, yPred) {
return tf.tidy(() => {
const zero = tf.scalar(0);
return tf.logicalAnd(yTrue.equal(zero), yPred.equal(zero))
.sum()
.cast('float32');
});
}
// TODO(cais): Use tf.metrics.falsePositiveRate when available.
function falsePositiveRate(yTrue, yPred) {
return tf.tidy(() => {
const fp = falsePositives(yTrue, yPred);
const tn = trueNegatives(yTrue, yPred);
return fp.div(fp.add(tn));
});
}
/**
* Draw a ROC curve.
*
* @param {tf.Tensor} targets The actual target labels, as a 1D Tensor
* object consisting of only 0 and 1 values.
* @param {tf.Tensor} probs The probabilities output by a model, as a 1D
* Tensor of the same shape as `targets`. It is assumed that the values of
* the elements are >=0 and <= 1.
* @param {number} epoch The epoch number where the `probs` values come
* from.
* @returns {number} Area under the curve (AUC).
*/
function drawROC(targets, probs, epoch) {
return tf.tidy(() => {
const thresholds = [
0.0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55,
0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.92, 0.94, 0.96, 0.98, 1.0
];
const tprs = []; // True positive rates.
const fprs = []; // False positive rates.
let area = 0;
for (let i = 0; i < thresholds.length; ++i) {
const threshold = thresholds[i];
const threshPredictions = utils.binarize(probs, threshold).as1D();
const fpr = falsePositiveRate(targets, threshPredictions).dataSync()[0];
const tpr = tf.metrics.recall(targets, threshPredictions).dataSync()[0];
fprs.push(fpr);
tprs.push(tpr);
// Accumulate to area for AUC calculation.
if (i > 0) {
area += (tprs[i] + tprs[i - 1]) * (fprs[i - 1] - fprs[i]) / 2;
}
}
ui.plotROC(fprs, tprs, epoch);
return area;
});
}
// Some hyperparameters for model training.
const epochs = 400;
const batchSize = 350;
const data = new WebsitePhishingDataset();
data.loadData().then(async () => {
await ui.updateStatus('Getting training and testing data...');
const trainData = data.getTrainData();
const testData = data.getTestData();
await ui.updateStatus('Building model...');
const model = tf.sequential();
model.add(tf.layers.dense(
{inputShape: [data.numFeatures], units: 100, activation: 'sigmoid'}));
model.add(tf.layers.dense({units: 100, activation: 'sigmoid'}));
model.add(tf.layers.dense({units: 1, activation: 'sigmoid'}));
model.compile(
{optimizer: 'adam', loss: 'binaryCrossentropy', metrics: ['accuracy']});
const trainLogs = [];
let auc;
await ui.updateStatus('Training starting...');
await model.fit(trainData.data, trainData.target, {
batchSize,
epochs,
validationSplit: 0.2,
callbacks: {
onEpochBegin: async (epoch) => {
// Draw ROC every a few epochs.
if ((epoch + 1) % 100 === 0 || epoch === 0 || epoch === 2 ||
epoch === 4) {
const probs = model.predict(testData.data);
auc = drawROC(testData.target, probs, epoch);
}
},
onEpochEnd: async (epoch, logs) => {
await ui.updateStatus(`Epoch ${epoch + 1} of ${epochs} completed.`);
trainLogs.push(logs);
ui.plotLosses(trainLogs);
ui.plotAccuracies(trainLogs);
}
}
});
await ui.updateStatus('Running on test data...');
tf.tidy(() => {
const result =
model.evaluate(testData.data, testData.target, {batchSize: batchSize});
const lastTrainLog = trainLogs[trainLogs.length - 1];
const testLoss = result[0].dataSync()[0];
const testAcc = result[1].dataSync()[0];
const probs = model.predict(testData.data);
const predictions = utils.binarize(probs).as1D();
const precision =
tf.metrics.precision(testData.target, predictions).dataSync()[0];
const recall =
tf.metrics.recall(testData.target, predictions).dataSync()[0];
const fpr = falsePositiveRate(testData.target, predictions).dataSync()[0];
ui.updateStatus(
`Final train-set loss: ${lastTrainLog.loss.toFixed(4)} accuracy: ${
lastTrainLog.acc.toFixed(4)}\n` +
`Final validation-set loss: ${
lastTrainLog.val_loss.toFixed(
4)} accuracy: ${lastTrainLog.val_acc.toFixed(4)}\n` +
`Test-set loss: ${testLoss.toFixed(4)} accuracy: ${
testAcc.toFixed(4)}\n` +
`Precision: ${precision.toFixed(4)}\n` +
`Recall: ${recall.toFixed(4)}\n` +
`False positive rate (FPR): ${fpr.toFixed(4)}\n` +
`Area under the curve (AUC): ${auc.toFixed(4)}`);
});
});