31:[] 2f:{"title":"Machine noise training","cells":[{"id":0,"value":"md`# Machine noise training`","pinned":false,"hidden":false,"mode":"ojs"},{"id":298,"value":"html`

`","pinned":false,"hidden":false,"mode":"ojs"},{"id":435,"value":"html`

`","pinned":false,"hidden":false,"mode":"ojs"},{"id":305,"value":" //return segments\n html`\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

From	To	Type
${0}	${formatSeconds(220*0.96)}	${labelDict[0]}
${formatSeconds(221*0.96)}	${formatSeconds(265*0.96)}	${labelDict[1]}
${formatSeconds(266*0.96)}	${formatSeconds(trainData.length*0.96)}	${labelDict[2]}

\n `\n \n ","pinned":false,"hidden":false,"mode":"ojs"},{"id":396,"value":"md`Número de muestras para train & test`","pinned":false,"hidden":false,"mode":"ojs"},{"id":381,"value":"{\n const total = trainTestInput.train.length + trainTestInput.test.length;\n const groupedTrain = _.groupBy(trainTestInput.train, d => d.class)\n const groupedTest = _.groupBy(trainTestInput.test, d => d.class)\n \n //return segments\n return html`\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n \n \n \n \n \n \n\n \n\n

Class	Train	Test	Total
${labelDict[0]}	${groupedTrain[0].length}	${groupedTest[0].length}	${groupedTrain[0].length+groupedTest[0].length}
${labelDict[1]}	${groupedTrain[1].length}	${groupedTest[1].length}	${groupedTrain[1].length+groupedTest[1].length}
${labelDict[2]}	${groupedTrain[2].length}	${groupedTest[2].length}	${groupedTrain[1].length+groupedTest[1].length}

\n `\n \n \n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":445,"value":"md`\n# Ficha técnica del registro\n* All audio is resampled to 16 kHz mono.\n* A spectrogram is computed using magnitudes of the Short-Time Fourier Transform with a window size of 25 ms, a window hop of 10 ms, and a periodic Hann window.\n* A mel spectrogram is computed by mapping the spectrogram to 64 mel bins covering the range 125-7500 Hz.\n* A stabilized log mel spectrogram is computed by applying log(mel-spectrum + 0.01) where the offset is used to avoid taking a logarithm of zero.\n* These features are then framed into non-overlapping examples of 0.96 seconds, where each example covers 64 mel bands and 96 frames of 10 ms each.\n\n* __Cada muesta de datos corespondiente a 0.96 seg (matriz de 96x64) es transormada a un vector de 128 dimensiones usando red neuronal VGGish__`","pinned":false,"hidden":false,"mode":"ojs"},{"id":400,"value":"{\n const annotated = trainData.map((d,i) => ({\n data: d,\n class: labelRule(i)\n }))\n return _.groupBy(annotated, d => d.class);\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":229,"value":"trainTestInput = {\n const annotated = trainData.map((d,i) => ({\n data: d,\n class: labelRule(i)\n }))\n const grouped = _.groupBy(annotated, d => d.class);\n \n const trainSet= []\n const testSet=[]\n \n _.each(grouped, (groupData, key) => {\n const items = _.shuffle(groupData)\n const numTrain = items.length *trainProportion;\n let i = 0;\n while(i< numTrain) {\n trainSet.push(items[i])\n i++\n }\n \n while(i < items.length) {\n testSet.push(items[i])\n i++\n }\n \n })\n \n return {train:trainSet,test: testSet}\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":22,"value":"model = {\n const model = tf.sequential();\n model.add(tf.layers.dense({units: 8, activation: 'relu', inputShape: [128]}));\n model.add(tf.layers.dense({units: 3, activation: 'softmax' }));\n\n // Prepare the model for training: Specify the loss and the optimizer.\n model.compile({loss: 'categoricalCrossentropy', optimizer: 'adam', metrics:[\"accuracy\"]});\n\n\n\n\n return model\n}\n","pinned":false,"hidden":false,"mode":"ojs"},{"id":164,"value":"trainModel = {\n trigger;\n // Generate some synthetic data for training.\n const xs = inputTensor;\n const ys = inputLabel;\n \n const options = {\n //batchSize: 32,\n epochs:50,\n //validationSplit: 0.10,\n //shuffle: true,\n callbacks: {\n onBatchBegin: onBatch,\n onEpochEnd: onEpochUpdate\n },\n //classWeight: {0:0.80, 1:0.20 }\n }\n \n // Train the model using the data.\n model.fit(xs, ys, options).then(() => {\n // Use the model to do inference on a data point the model hasn't seen before:\n // Open the browser devtools to see the output\n //model.predict(tf.tensor2d([5], [1, 1])).print();\n console.log(\"Done\")\n });\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":455,"value":"viewof trigger = html``","pinned":false,"hidden":false,"mode":"ojs"},{"id":51,"value":"mutable log = null","pinned":false,"hidden":false,"mode":"ojs"},{"id":260,"value":"evaluate = {\n log;\n const evalData = tf.tensor(trainTestInput.test.map(d=>d.data));\n const evalLabels = tf.oneHot(trainTestInput.test.map(d => d.class),3)\n \n const result = model.evaluate(evalData, evalLabels)\n \n //return result.toString()\n \n const predict = model.predict(evalData);\n const predictedLabels = predict.argMax(1).dataSync();\n const correctLabels = trainTestInput.test.map(d => d.class); \n const resultDetail= _.map(correctLabels,(d,i) => ({label:d, predicted: predictedLabels[i], correct: d == predictedLabels[i]}))\n const result2= _.map(_.groupBy(resultDetail, d => d.label), (items,label) =>({label:label, correct: items.filter(d => d.correct).length, incorrect: items.filter(d => !d.correct).length}))\n \n return html`${result2.map(d =>`

${labelDict[d.label]} ${d.correct}/${d.correct+d.incorrect}

`)}`\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":170,"value":"saveModel = {\n //model.save('downloads://model_noise')\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":11,"value":"labelDict = [\n \"normal\",\n \"descarga\",\n \"cierre válvula\"\n]","pinned":false,"hidden":false,"mode":"ojs"},{"id":27,"value":"inputTensor = tf.tensor(trainTestInput.train.map(d=>d.data))","pinned":false,"hidden":false,"mode":"ojs"},{"id":45,"value":"inputLabel = tf.oneHot(trainTestInput.train.map(d => d.class),3)","pinned":false,"hidden":false,"mode":"ojs"},{"id":247,"value":"inputLabel.shape","pinned":false,"hidden":false,"mode":"ojs"},{"id":236,"value":"trainProportion =0.8","pinned":false,"hidden":false,"mode":"ojs"},{"id":71,"value":"labelRule = (i) => {\n if (i > 265) {\n return 2\n } else if (i > 219) {\n return 1\n } else{\n return 0\n }\n\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":41,"value":"trainData[312]","pinned":true,"hidden":false,"mode":"ojs"},{"id":64,"value":"onBatch = (a,b) => {\n console.log(a,b)\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":54,"value":"onEpochUpdate= (a,b) => {\n debugger;\n mutable log = `${a} ${b.loss} ${b.acc}`\n console.log(a,b)\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":78,"value":"prediction = loadedModel.predict(tf.tensor(testData))","pinned":false,"hidden":false,"mode":"ojs"},{"id":91,"value":"displayPrediction = (p) => {\n const data = p.dataSync();\n const rows = p.shape[0];\n const cols = p.shape[1];\n \n const out = [];\n\n \n for (let row of _.range(rows)) {\n const item = _.slice(data, row * cols, row * cols + cols);\n out.push(item)\n }\n \n \n return out\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":183,"value":"buildSegments = (data) => {\n debugger;\n const segments = [];\n \n let startSec = data[0].sec;\n let currentSec = data[0].sec;\n let currentClass = data[0].class;\n \n data.forEach(d => {\n const thisClass = d.class;\n if (thisClass !== currentClass) {\n segments.push({\n from: startSec,\n to: currentSec,\n state: currentClass\n })\n startSec = d.sec;\n currentSec = d.sec;\n currentClass = d.class;\n }\n currentSec = d.sec;\n })\n segments.push({\n from: startSec,\n to: currentSec,\n state: currentClass\n })\n \n return segments\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":151,"value":"prediction.argMax(1).dataSync()","pinned":false,"hidden":false,"mode":"ojs"},{"id":130,"value":"labelDict[1]","pinned":false,"hidden":false,"mode":"ojs"},{"id":95,"value":"loadedModel = tf.loadModel('https://raw.githubusercontent.com/elaval/datasets/master/model-noise/model_noise.json')","pinned":false,"hidden":false,"mode":"ojs"},{"id":81,"value":"testData= _.slice(testDataStr.split(\"\\n\").map(d => d.split(\"\\t\")),0,310)","pinned":false,"hidden":false,"mode":"ojs"},{"id":18,"value":"testDataStr = d3.text(\"https://raw.githubusercontent.com/elaval/datasets/master/test.tsv\")","pinned":false,"hidden":false,"mode":"ojs"},{"id":34,"value":"trainData = _.slice(trainDataStr.split(\"\\n\").map(d => d.split(\"\\t\")),0,312)","pinned":false,"hidden":false,"mode":"ojs"},{"id":14,"value":"trainDataStr = d3.text(\"https://raw.githubusercontent.com/elaval/datasets/master/train.tsv\")","pinned":false,"hidden":false,"mode":"ojs"},{"id":427,"value":"formatSeconds(200)","pinned":false,"hidden":false,"mode":"ojs"},{"id":406,"value":"formatSeconds = (d) => {\n return moment.utc(d*1000).format('mm:ss.S')\n}\n","pinned":false,"hidden":false,"mode":"ojs"},{"id":404,"value":"moment= require(\"moment\")","pinned":false,"hidden":false,"mode":"ojs"},{"id":2,"value":"tf = require('@tensorflow/tfjs@0.13.0')","pinned":false,"hidden":false,"mode":"ojs"},{"id":7,"value":"d3 = require(\"d3\")","pinned":false,"hidden":false,"mode":"ojs"},{"id":9,"value":"_ = require(\"lodash\")","pinned":false,"hidden":false,"mode":"ojs"}],"files":"$W31","stdlib":"1","titling":"auto"}