-
Notifications
You must be signed in to change notification settings - Fork 41
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
0a8b94b
commit ce44dcc
Showing
5 changed files
with
198 additions
and
29 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,154 @@ | ||
| { | ||
| "cells": [ | ||
| { | ||
| "language": "markdown", | ||
| "source": [ | ||
| "# A Non-linear regression model for Used Car price Prediction\nCreation of a simple non linear regression model for used Audi car price from the given data set\n\n[https://medium.com/@swathylenjini/non-linear-regression-for-used-car-price-prediction-model-tensorflow-js-bcce05f970c5](https://medium.com/@swathylenjini/non-linear-regression-for-used-car-price-prediction-model-tensorflow-js-bcce05f970c5)\n\n### SWATHY L" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "markdown", | ||
| "source": [ | ||
| "# Prerequisites\n\n* Download audi.csvs from https://www.kaggle.com/adityadesai13/used-car-dataset-ford-and-mercedes?select=audi.csv\n* install `@tensorflow/tfjs-node`, `@tensorflow/tfjs-vis`" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "typescript", | ||
| "source": [ | ||
| "const tf = require('@tensorflow/tfjs-node') as typeof import('@tensorflow/tfjs-node');\nconst tfvis = require('@tensorflow/tfjs-vis') as typeof import('@tensorflow/tfjs-vis');" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "async function plot(pointsArray, featureName) {\n tfvis.render.scatterplot(\n { name: `${featureName} vs Car Price` },\n { values: [pointsArray], series: ['original'] },\n {\n xLabel: featureName,\n yLabel: 'Price',\n }\n )\n}" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "markdown", | ||
| "source": [ | ||
| "# Case 1: Lets evaluate the model using mileage as a feature." | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "// Import data\nconst audi_data = tf.data.csv('file://<fully qualified file path>/audi.csv');\n// Extract x and y values to plot in graph\nconst dataSet = audi_data.map(record => ({\n x: record.mileage,\n y: record.price,\n}));\nconst points = await dataSet.toArray();\nplot(points, 'Mileage');" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "function normalise(tensor) {\n const min = tensor.min();\n const max = tensor.max();\n const normalisedTensor = tensor.sub(min).div(max.sub(min));\n return {\n tensor: normalisedTensor,\n min,\n max\n };\n}\n" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "//Shuffle\ntf.util.shuffle(points);\n// Extract Features (inputs)\nconst featureValues = points.map(p => p.x);\nconst featureTensor = tf.tensor2d(featureValues, [featureValues.length, 1]);\n// Extract Labels (outputs)\nconst labelValues = points.map(p => p.y);\nconst labelTensor = tf.tensor2d(labelValues, [labelValues.length, 1]);\n// Normalise features and labels\nconst normalisedFeature = normalise(featureTensor);\nconst normalisedLabel = normalise(labelTensor);\n" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "const [trainingFeatureTensor, testingFeatureTensor] = tf.split(normalisedFeature.tensor, 2);\nconst [trainingLabelTensor, testingLabelTensor] = tf.split(normalisedLabel.tensor, 2);" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "function createModel() {\n model = tf.sequential();\n model.add(tf.layers.dense({\n units: 1,\n useBias: true,\n activation: 'sigmoid',\n inputDim: 1\n }));\n model.compile({\n loss: 'meanSquaredError',\n optimizer: tf.train.adam()\n });\n return model;\n}" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "//Create Model\nconst model = createModel();" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "async function trainModel(model, trainingFeatureTensor, trainingLabelTensor, trainingLabel) {\n const { onBatchEnd, onEpochEnd } = tfvis.show.fitCallbacks({ name: trainingLabel }, ['loss']);\n return model.fit(trainingFeatureTensor, trainingLabelTensor, {\n batchSize: 32,\n epochs: 20,\n validationSplit: 0.2,\n callbacks: {\n onEpochEnd,\n }\n });\n}" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "//Train Model\nconst result = await trainModel(model, trainingFeatureTensor, trainingLabelTensor, \"Training Performance\");" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "const trainingLoss = result.history.loss.pop();\nconsole.log(`Training loss: ${trainingLoss}`);\nconst validationLoss = result.history.val_loss.pop();\nconsole.log(`Validation loss: ${validationLoss}`);" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "//Evaluate Model\nconst lossTensor = model.evaluate(testingFeatureTensor, testingLabelTensor);\nconst loss = await lossTensor.dataSync();\nconsole.log(`Testing loss: ${loss}`);" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "markdown", | ||
| "source": [ | ||
| "# Case 2: Let's evaluate the model using mpg as a feature." | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "// Import data\nconst audi_data = tf.data.csv('file://<fully qualified file path>/audi.csv');\n// Extract x and y values to plot in graph\nconst dataSet = audi_data.map(record => ({\n x: record.mpg,\n y: record.price,\n}));\nconst points = await dataSet.toArray();\nplot(points, 'MPG');\n" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "markdown", | ||
| "source": [ | ||
| "# Case 3: Lets evaluate model using both mileage and mpg as features for price prediction model. " | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "function createModel() {\n model = tf.sequential();\n model.add(tf.layers.dense({\n units: 10,\n useBias: true,\n activation: 'sigmoid',\n inputDim: 2,\n }));\n model.add(tf.layers.dense({\n units: 1,\n useBias: true,\n activation: 'sigmoid',\n }));\n model.compile({\n loss: 'meanSquaredError',\n optimizer: tf.train.adam()\n });\n return model;\n}" | ||
| ], | ||
| "outputs": [] | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "const dataSet = audi_data.map((record) => ({\n x1: record.mpg,\n x2: record.mileage,\n y: record.price\n}));\nconst points = await dataSet.toArray();\n//Shuffle\ntf.util.shuffle(points);\n// Extract and normalise features\nconst normalisedFeature1 = normalise(tf.tensor2d(points.map((p) => [p.x1]))).tensor.dataSync();\nconst normalisedFeature2 = normalise(tf.tensor2d(points.map((p) => [p.x2]))).tensor.dataSync();\nconst normFeatureInput = [];\nnormalisedFeature1.forEach((item, i, self) => { normFeatureInput.push([normalisedFeature1[i], normalisedFeature2[i]]) })\n// Extract and normalise Labels (outputs)\nconst labelValues = points.map(p => p.y);\nconst labelTensor = tf.tensor2d(labelValues, [labelValues.length, 1]);\nconst normalisedLabel = normalise(labelTensor);\n//Training and testing data sets\nconst [trainingFeatureTensor, testingFeatureTensor] = tf.split(normFeatureInput, 2);\nconst [trainingLabelTensor, testingLabelTensor] = tf.split(normalisedLabel.tensor, 2);\n//Create Model\nconst model = createModel();\n//Train Model\nconst result = await trainModel(model, trainingFeatureTensor, trainingLabelTensor, \"Training Performance (case 3)\");\nconst trainingLoss = result.history.loss.pop();\nconsole.log(`Training loss: ${trainingLoss}`);\nconst validationLoss = result.history.val_loss.pop();\nconsole.log(`Validation loss: ${validationLoss}`);\n//Evaluate Model\nconst lossTensor = model.evaluate(testingFeatureTensor, testingLabelTensor);\nconst loss = await lossTensor.dataSync();\nconsole.log(`Testing loss: ${loss}`);" | ||
| ], | ||
| "outputs": [], | ||
| "metadata": { | ||
| "outputCollapsed": true | ||
| } | ||
| }, | ||
| { | ||
| "language": "javascript", | ||
| "source": [ | ||
| "tfvis.show.history({ name: 'Case 3 Training History' }, result, ['loss'])" | ||
| ], | ||
| "outputs": [] | ||
| } | ||
| ] | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters