PL SQLDay Machine Learning- Hands on ML.NET.pptx
•Download as PPTX, PDF•
0 likes•260 views
Slides for the hands-on-lab workshop about Machine Learning with ML.NET at SQL Day Poland 2021
Recommended
Recommended
More Related Content
What's hot
What's hot (20)
Similar to PL SQLDay Machine Learning- Hands on ML.NET.pptx
Similar to PL SQLDay Machine Learning- Hands on ML.NET.pptx (20)
More from Luis Beltran
More from Luis Beltran (20)
Recently uploaded
Recently uploaded (20)
PL SQLDay Machine Learning- Hands on ML.NET.pptx
- 1. SQLDay 2021
- 2. Machine Learning: Hands on ML.NET Luis Beltrán SQLDay 2021
- 3. Luis Beltrán • Researcher - Tomas Bata University in Zlín, Czech Republic. • Lecturer - Tecnológico Nacional de México en Celaya, Mexico. • Xamarin, Azure and Artificial Intelligence @darkicebeam luis@luisbeltran.mx
- 4. AGENDA • Machine Learning • ML.NET • ML Workflow with Ml.NET • Data • Train • Evaluate • Save model • Consume model • Deep Learning • MLOps SQLDay 2021
- 5. Objectives • Build, train, evaluate, and consume machine learning algorithms in your .NET apps using ML.NET. • Understand how TensorFlow (and ONNX) models can be integrated into a pipeline for deep learning. • Set up model lifecycle automation using MLOps. SQLDay 2021
- 7. Artificial Intelligence The ability of a computer to perform tasks commonly associated with intelligent beings (reason, discover meaning, generalize, learn from past experience) Artificial Intelligence • Typically starts as rule or logic-based system • Traditional AI techniques can be difficult to scale SQLDay 2021
- 8. Machine Learning Machine Learning Getting computers to make predictions without being explicitly programmed • Computers find patterns in data and learn from experience to act on new data • Used to solve problems that are difficult or impossible to solve with rules-based programming SQLDay 2021
- 9. Machine Learning Bread or not bread? Bread Not bread SQLDay 2021
- 10. Artificial Intelligence vs. Machine Learning Artificial Intelligence Machine Learning Rules Data Data Answers Answers Rules SQLDay 2021
- 11. Deep learning Deep Learning Subset of ML based on artificial neural networks which imitate the way the human brain learns, thinks, and processes data. • Neural networks form many layers • Scenarios include image classification, object detection, speech recognition, NLP SQLDay 2021
- 12. AI + ML + Deep learning Artificial Intelligence Machine Learning Deep Learning SQLDay 2021
- 13. Mapping business problems to ML Tasks What problem are you looking to solve? Find outliers • Anomaly detection Predict a number • Regression • Forecasting Find relationships • Clustering Categorize items • Binary classification • Multiclass classification •Image classification Make suggestions •Recommendation SQLDay 2021
- 14. Machine Learning Workflow Prepare the data Evaluate Train Deploy Model Training Model Consumption Inferencing Get the data SQLDay 2021
- 15. Get and prepare the data Data Source Pipeline Environment Data exploration SQLDay 2021
- 16. Prepare the data Data exploration Visualization Data cleaning SQLDay 2021
- 17. Train the model Previous Grade (A-F) Hours Studied Pass B 5 Y D 2 N A 20 Y Features Label / Target F(PreviousGrade, HoursStudied) = Pass Model SQLDay 2021
- 19. Deploy Model files Deployment targets SQLDay 2021
- 20. Automated Machine Learning Get the data Prepare the data Train Evaluate Inferencing Deploy Model Training Model Consumption SQLDay 2021
- 23. Machine learning landscape At Microsoft Azure Cognitive Services Azure Machine Learning WinML SQLDay 2021
- 24. Machine learning landscape At Microsoft Training custom models Model consumption Requires ML knowledge ML.NET Yes Yes - ML.NET, TensorFlow, ONNX No Azure Cognitive Services Limited to some services Yes – consume via API/SDK No Azure ML Yes Yes – register models & consume via web service Somewhat WinML No Yes - ONNX No SQLDay 2021
- 26. An open source and cross-platform machine learning framework for .NET Windows Linux macOS SQLDay 2021
- 27. Built for .NET Can use existing C# and F# skills to integrate ML into .NET apps Data science & ML experience not required Developers SQLDay 2021
- 28. ML.NET Tooling + AutoML ML.NET API (Microsoft.ML) AutoML.NET API (Microsoft.ML.AutoML) Model Builder ML.NET CLI SQLDay 2021
- 29. Model Builder & ML.NET CLI • Easily build custom ML models with AutoML • Generates code for training and consumption • Model Builder • Currently in Visual Studio only (ships with VS 16.6) • Integration with Azure ML (image classification) • ML.NET CLI • Cross platform SQLDay 2021
- 30. Supported ML tasks in ML.NET Classification Regression Image classification Anomaly detection Forecasting Object detection Clustering Recommendation Ranking SQLDay 2021
- 31. Integration with other ML tech @ Microsoft Azure Cognitive Services (Custom Vision) Train Image classification or object detection Consume In .NET app using ML.NET Export To ONNX Azure Cognitive Services – Custom Vision SQLDay 2021
- 32. Integration with other ML tech @ Microsoft Train Using Azure AutoML Start in Model Builder Choose Scenario, Training Environment, & Data Consume In .NET app using Model Builder & ML.NET Azure Machine Learning Azure ML Model Builder in VS SQLDay 2021
- 33. Integration with other ML tech @ Microsoft Train Using ML.NET Consume With WinML in Windows Desktop Apps WinML Export To ONNX WinML SQLDay 2021
- 34. • Want to stay in .NET ecosystem for Machine Learning • Don’t want to worry about low-level complexities of ML • Want to train a custom model • Want to consume a pre-trained model When should you use ML.NET? When you… SQLDay 2021
- 35. ML Workflow with Ml.NET SQLDay 2021
- 36. • We will create a web app that allows users to input in data about a taxi trip and returns how much they will pay (taxi fare). • Regression task (value prediction scenario) • App details: • Train regression model in .NET core console app with given dataset • Consume model in ASP.NET Core web app Problem to solve: Taxi fare prediction SQLDay 2021
- 37. • Most scenarios • Microsoft.ML • Forecasting & anomaly detection scenarios : • Microsoft.ML.TimeSeries • Recommendation scenario: • Microsoft.ML.Recommender • Database loader • System.Data.SqlClient ML.NET NuGet Packages • Consuming ONNX models: • Microsoft.ML.ONNXTransformer (+ Microsoft.ML.ImageAnalytics for object detection) • Consuming TensorFlow models: • Microsoft.ML.TensorFlow + SciSharp.TensorFlow.Redist (+ Microsoft.ML.ImageAnalytics for image classification) • Train custom image classification models: • Microsoft.ML.Vision + Microsoft.ML.ImageAnalytics + SciSharp.TensorFlow.Redist SQLDay 2021
- 38. • MLContext = starting point for all ML.NET operations • Provides ways to create components for • Data preparation • Feature engineering • Training • Prediction • Model evaluation • Logging • Execution control • Seeding MLContext SQLDay 2021
- 39. Task 1. Add the Microsoft.ML NuGet package to your console project 2. Initialize a new MLContext in your console app SQLDay 2021
- 40. Data in ML.NET represented as IDataView IDataView High-dimensional Lazy + memory efficient Immutable SQLDay 2021
- 41. • DataViewSchema = Data schema of IDataView = set of columns, their names, types, & other annotations • Before loading data, must define how schema of data will look (column names & column types) • Use class definitions to define IDV schemas Data schema Class definition of schema Dataset Label SepalLength SepalWidth PetalLength PetalWidth Iris-setosa 5.1 3.5 1.4 0.2 Iris-versicolor 7.0 3.2 4.7 1.4 Iris-setosa 4.9 3.0 1.5 0.1 … IDataView SQLDay 2021
- 42. File loaders • Load data from sources like text, binary, and image files to IDV • Can load from single or multiple files • Supported: • Text: .csv, .tsv, .txt • Images: .png, .jpg, .bmp Data loaders & sources Database loaders • Load and train data directly from relational database • Supports: • SQL Server, Azure SQL Database, Oracle, SQLite, PostgreSQL, Progress, IBM DB2, + many more Other sources • Load from Enumerable (in- memory collections) • Supports: • JSON/XML • Everything else SQLDay 2021
- 43. Task 1. Create class for Model Input based on the provided taxi trip dataset 2. Load data from file to IDataView SQLDay 2021
- 44. Preparing your data Filter data Convert data types Normalize the data Split data Feature engineering SQLDay 2021
- 45. IEstimator and ITransformer IEstimator ITransformer SQLDay 2021
- 46. Normalization • Min-Max • Binning • Mean variance Missing Values • Indicate • Replace ColumnMapping • Concatenate • Copy columns • Drop columns Type Conversion • Convert type • Map value to key • Hash Text Transforms • Featurize text • Remove stop words • N-grams • Word bags Data transforms SQLDay 2021
- 47. Algorithms / Trainers Trainer = Algorithm + Task Example: Stochastic Dual Coordinated Ascent (SDCA) Binary classification Multi-class classification Regression SdcaNonCalibratedMulti classTrainer SdcaRegressionTrainer SdcaNonCalibrated BinaryTrainer Algorithm Task Trainer SQLDay 2021
- 48. IEstimatorChain = Collection of Data Transforms + Algorithms Training pipeline IDataView IEstimatorChain Model Drop columns Normalize Naïve Bayes Algorithm SQLDay 2021
- 49. Pipeline executed when Fit() method is called Fit() the model ITransformer model = pipeline.Fit(trainingData) SQLDay 2021
- 50. Task 1. Split data into train and test datasets 2. Add data transformations to the pipeline 3. Choose an algorithm and add to the pipeline 4. Train the model SQLDay 2021
- 51. Evaluation metrics ML Task Most common evaluation metric Look for Classification Binary: Accuracy Multi-class: Micro-Accuracy Closer to 1.0, the better the quality Regression R-Squared Closer to 1.0, the better the quality Recommendation R-Squared Closer to 1.0, the better the quality Clustering Average Distance Values closer to 0 are better. Ranking Discounted Cumulative Gains Higher values are better Anomaly detection Area Under ROC Curve Values closer to 1 are better. SQLDay 2021
- 52. Underfitting & Overfitting a model Underfitting Model is too simple and can’t capture the underlying trend of the data Overfitting Model doesn’t generalize well from training data to unseen data To prevent: • Feature engineering • Remove noise from data • Try different algorithms To prevent: • More training data • Remove features • Cross validation SQLDay 2021
- 53. • Training and model evaluation technique • Folds the data into n-partitions and trains multiple algorithms on these partitions • Improves robustness by holding out data from training process Cross validation Partition 1 Partition 2 Partition 3 Partition 4 Partition 5 SQLDay 2021
- 54. • Global and local explanations • Global = entire model (What features does the model give more importance to?) • Local = individual predictions (Why was Bob rejected for a loan?) • Techniques: Model explainability Permutation Feature Importance (PFI) • Used for Classification and Regression models • Shuffles data one feature at a time and calculates how much the performance metric of interest decreases; the larger the change, the more important the feature Feature Contribution Calculation (FCC) • Used for Classification and Regression models • Shows which features are most influential for a model’s prediction on a particular and individual data sample SQLDay 2021
- 55. • Provide more training data • Filter missing values and outliers • Feature engineering • Select different features • Choose a different algorithm • Tune algorithm hyperparameters • Cross validation Improving your model SQLDay 2021
- 56. • Use AutoML to speed up the experimentation process • Use Model Builder in VS or cross-platform ML.NET CLI Tooling + AutoML ML Task Tooling Local / Azure AutoML Text-based classification Model Builder, CLI Local Value prediction (Regression) Model Builder, CLI Local Image classification Model Builder Local + Azure Recommendation Model Builder, CLI Local SQLDay 2021
- 57. Task 1. Evaluate your model and print out the metrics 2. Optional: Try training with different algorithms to see if your evaluation metrics change SQLDay 2021
- 58. ML.NET Model ML.NET Model = MLModel.zip Serialized zip file which contains data schemas, data transforms, and algorithms SQLDay 2021
- 59. Task 1. Save your model SQLDay 2021
- 60. 1. Create model output schema How to consume model in ML.NET Task Model Output Binary classification Predicted Label: Class predicted by model (true or false) Score: Positive score = true, negative score = false Probability: Probability of having true as label Multiclass classification Predicted Label: Class predicted by model Score (vector): Scores of all classes; highest score = predicted label Regression Score: Predicted value Recommendation Score: Predicted rating Clustering Predicted Label: Closest cluster’s index predicted by model Score: Distances of data point to clusters’ centroid Ranking Score: Predicted rank Anomaly detection <Alert (Boolean), Raw Score, P-value (likelihood of anomaly)> OR Predicted Label: Anomaly vs. not anomaly predicted by model Score: Likelihood of anomaly Forecasting Forecasting values Confidence lower bounds Confidence upper bounds SQLDay 2021
- 61. 2. Load your model How to consume model in ML.NET SQLDay 2021
- 62. How to consume model in ML.NET • Prediction Engine = convenience API for making single predictions Make single predictions Prediction Engine • Prediction Engine not thread- safe • Use dependency injection + Prediction Engine Pool in multi- threaded apps (e.g. web apps and services) • Creates ObjectPool of PredictionEngine objects for application use Make single predictions scalable Prediction Engine Pool • Takes in data, makes the transformations (such as, making predictions), and outputs the data • Can load unknown data into IDataView, use Transform to predict, receive IDataView of predicted values, and use GetColumn to get the Prediction column Make batch predictions Transform 3. Choose one of the below: SQLDay 2021
- 63. Model deployment Desktop Web Mobile SQLDay 2021
- 64. Task 1. Load the model from a file to the web app 2. Create a Prediction Engine 3. Use the model and prediction engine to make predictions on new sample data (e.g. consume the model) SQLDay 2021
- 66. What are Neural Networks? SQLDay 2021
- 67. Deep Learning • Deep learning is a subfield of Machine Learning concerned with algorithms inspired by the structure and function of the brain called artificial neural networks. • It is exceptionally effective in discovering patterns. • Algorithms learn through a multi-layered hierarchy. • If you supply the system with tons of information, it will begin to understand and respond in helpful ways. SQLDay 2021
- 68. Deep learning has an inbuilt automatic multi stage feature learning process that learns rich hierarchical representations (i.e. features). Low-level features Mid-level features Output (e.g. exterior, interior) High-level features Trainable Classifier SQLDay 2021
- 69. • Image Pixel Edge Texture Motif Part Object • Text Character Word Word-group Clause Sentence Story • Each module in Deep Learning transforms its input representation into a higher-level one, in a way similar to human cortex. Low Level Features Mid Level Features Output High Level Features Trainable Classifier Input SQLDay 2021
- 70. Convolutional Layers Filter 1 1 1 1 1 1 0.015686 0.015686 0.011765 0.015686 0.015686 0.015686 0.015686 0.964706 0.988235 0.964706 0.866667 0.031373 0.023529 0.007843 0.007843 0.741176 1 1 0.984314 0.023529 0.019608 0.015686 0.015686 0.015686 0.011765 0.101961 0.972549 1 1 0.996078 0.996078 0.996078 0.058824 0.015686 0.019608 0.513726 1 1 1 0.019608 0.015686 0.015686 0.015686 0.007843 0.011765 1 1 1 0.996078 0.031373 0.015686 0.019608 1 0.011765 0.015686 0.733333 1 1 0.996078 0.019608 0.019608 0.015686 0.015686 0.011765 0.984314 1 1 0.988235 0.027451 0.015686 0.007843 0.007843 1 0.352941 0.015686 0.823529 1 1 0.988235 0.019608 0.019608 0.015686 0.015686 0.019608 1 1 0.980392 0.015686 0.015686 0.015686 0.015686 0.996078 1 0.996078 0.015686 0.913726 1 1 0.996078 0.019608 0.019608 0.019608 0.019608 1 1 0.984314 0.015686 0.015686 0.015686 0.015686 0.952941 1 1 0.992157 0.019608 0.913726 1 1 0.988235 0.019608 0.019608 0.019608 0.039216 0.996078 1 0.015686 0.015686 0.015686 0.015686 0.996078 1 1 1 0.007843 0.019608 0.898039 1 1 0.988235 0.019608 0.015686 0.019608 0.968628 0.996078 0.980392 0.027451 0.015686 0.019608 0.980392 0.972549 1 1 1 0.019608 0.043137 0.905882 1 1 1 0.015686 0.035294 0.968628 1 1 0.023529 1 0.792157 0.996078 1 1 0.980392 0.992157 0.039216 0.023529 1 1 1 1 1 0.992157 0.992157 1 1 0.984314 0.015686 0.015686 0.858824 0.996078 1 0.992157 0.501961 0.019608 0.019608 0.023529 0.996078 0.992157 1 1 1 0.933333 0.003922 0.996078 1 0.988235 1 0.992157 1 1 1 0.988235 1 1 1 1 0.015686 0.74902 1 1 0.984314 0.019608 0.019608 0.031373 0.984314 0.023529 0.015686 0.015686 1 1 1 0 0.003922 0.027451 0.980392 1 0.019608 0.023529 1 1 1 0.019608 0.019608 0.564706 0.894118 0.019608 0.015686 0.015686 1 1 1 0.015686 0.015686 0.015686 0.05098 1 0.015686 0.015686 1 1 1 0.047059 0.019608 0.992157 0.007843 0.011765 0.011765 0.015686 1 1 1 0.015686 0.019608 0.996078 0.023529 0.996078 0.019608 0.015686 0.243137 1 1 0.976471 0.035294 1 0.003922 0.011765 0.011765 0.015686 1 1 1 0.988235 0.988235 1 0.003922 0.015686 0.019608 0.019608 0.027451 1 1 0.992157 0.223529 0.662745 0.011765 0.011765 0.011765 0.015686 1 1 1 0.015686 0.023529 0.996078 0.011765 0.011765 0.015686 0.015686 0.011765 1 1 1 1 0.035294 0.011765 0.011765 0.011765 0.015686 1 1 1 0.015686 0.015686 0.964706 0.003922 0.996078 0.007843 0.019608 0.011765 0.054902 1 1 0.988235 0.007843 0.011765 0.011765 0.015686 0.011765 1 1 1 0.015686 0.015686 0.015686 0.023529 1 0.007843 0.007843 0.015686 0.015686 0.960784 1 0.490196 0.015686 0.015686 0.015686 0.007843 0.027451 1 1 1 0.011765 0.011765 0.043137 1 1 0.023529 0.003922 0.007843 0.023529 0.980392 0.976471 0.039216 0.019608 0.007843 0.019608 0.015686 1 1 1 1 1 1 1 1 1 0 1 0 1 -4 1 0 1 0 Input Image Convoluted Image SQLDay 2021
- 71. Convolution Input Image Convolved Image (Feature Map) a b c d e f g h i j k l m n o p w1 w2 w3 w4 Filter h1 h2 ℎ1 = 𝑓 𝑎 ∗ 𝑤1 + 𝑏 ∗ 𝑤2 + 𝑒 ∗ 𝑤3 + 𝑓 ∗ 𝑤4 ℎ2 = 𝑓 𝑏 ∗ 𝑤1 + 𝑐 ∗ 𝑤2 + 𝑓 ∗ 𝑤3 + 𝑔 ∗ 𝑤4 SQLDay 2021
- 72. Lower Level to More Complex Features Input Image Layer 1 Feature Map Layer 2 Feature Map w1 w2 w3 w4 w5 w6 w7 w8 Filter 1 Filter 2 SQLDay 2021
- 73. Pooling • Max pooling: reports the maximum output within a rectangular neighborhood. • Average pooling: reports the average output of a rectangular neighborhood. 1 3 5 3 4 2 3 1 3 1 1 3 0 1 0 4 MaxPool with 2X2 filter with stride of 2 Input Matrix Output Matrix 4 5 3 4 SQLDay 2021
- 74. Convolutional Neural Network Feature Extraction Architecture 64 64 128 128 256 256 256 512 512 512 512 512 512 Filter Max Pool Fully Connected Layers Living Room Bed Room Kitchen Bathroom Outdoor Maxpool Output Vector SQLDay 2021
- 75. Training Deep Learning Models in ML.NET Architectures • MobileNet • Inception • Resnet SQLDay 2021
- 76. Deep learning in ML.NET • Model Training Image Classification API • Train custom image classification models via Image Classification API • Uses transfer learning • Built on TensorFlow.NET • Can use local GPU for training • Model Consumption ML.NET API • Consume pre-trained TensorFlow and ONNX models Model Training Model Builder in VS • Train custom image classification models • Can train locally or in Azure (Azure ML) SQLDay 2021
- 77. Task • Task: Local image training with Image Classification API SQLDay 2021
- 78. Input Model Solution structure Nuget Packages Output Model (Prediction) SQLDay 2021
- 79. Main libraries Paths PrepareSet: Loading input images for training, validation, and testing SQLDay 2021
- 80. Display information to the Console Input data (images) SQLDay 2021
- 82. Main program Loading data for supervised learning (images include tags) Training and Validation sets Load pipeline: Images loaded in memory Training options: ImageClassificationTrainer chosen, based on the InceptionV3 architecture Training pipeline: Trying to predict a category Both pipelines are combined SQLDay 2021
- 83. Perform training Model precision is validated using validation dataset Model Metrics calculated Test the classification model using the new images Prepare new images for validation Export the model Consume the model SQLDay 2021
- 84. ConsumingModel Load a previously trained classification model and prepare test images that were not used before in the training and validation stages ClassifyImages: Test the model with new images SQLDay 2021
- 86. Validation Precision metrics Model validation results: Image Actual damage class Prediction SQLDay 2021
- 87. Image classification results ML Model exported as zip file SQLDay 2021
- 89. You can load a pre-trained TensorFlow model to integrate it in an ML.NET pipeline: SQLDay 2020
- 90. Task Task: Analyze sentiment of movie reviews using a pre-trained TensorFlow model in ML.NET SQLDay 2021
- 92. Machine Learning Operations (MLOps) applies DevOps principles & practices (e.g. continuous integration, delivery, and deployment) to the ML process What is ML Ops? SQLDay 2021
- 94. Task Task: Build a project that trains, tests, and deploys a model. SQLDay 2021
- 95. ONNX SQLDay 2020
- 96. Thank you for your attention luis@luisbeltran.mx
- 97. SQLDay 2021
Editor's Notes
- AI = machines imitating human abilities and behaviors
- Classification: Categorize e-mail as spam or not spam Automatically add labels to GitHub issues Predict of sentiment of a comment is positive or negative Diagnose if a patient has a disease or not Categorize hotel reviews into location, price, cleanliness, etc. Regression Predict the price of a house based on house features Predict sales of a product based on advertising budgets Predict number of goals a player will scare in upcoming match based on previous perf Recommendation Recommend products to shoppers Recommends movies or songs to users Forecasting Predict future sales based on past sales Predictive maintenance Weather forecasting Product demand forecasting Anomaly detection Identify fraudulent transactions Alert when there are predicted network intrusions Find abnormal cluster of patients Clustering Customer segmentation Ranking Search results ranking
- Data Source Database Files Pipeline Streaming Batch Environment On-Premises Cloud Exploratory Data Analysis Visualization Cleanup Validation
- Date exploration Distribution of the data What are the ranges (age?) Data cleaning Remove null values Visualization Try to identify patterns in your data
- Training a model consists of providing data examples to an algorithm that learns the mappings from the inputs to outputs. This is like a function that maintains an internal state. You can then reuse this model to make predictions on new data.
- Metrics – These are there to see whether your model is performing accurately Explainability – How easy is it to interpret the model Training effort – Does it take too long to prepare the data for the model or does it take too long to train the model? If it takes a week to gather the data for a particular model maybe it might be best to trade accuracy for ease of use and training.
- Deployment can mean many things Deployment can mean uploading a file This also means that you can then deploy the model to multiple deployment targets
- ML is very experimental / trial and error Preparing the data Training Evaluating is an art as much as a science You can’t just say, “classification, use Naïve Bayes”. Your data has a lot to do with what works and doesn’t work.
- Scikit Learn – Traditional / Classical Machine Learning TensorFlow – ML Deep Learning PyTorch – ML Deep Learning Keras – Higher level API for building ML that can use a variety of backends to actually build the models and computations
- Classification – sentiment analysis, Categorizing e-mail as spam or not spam, classify law documents Regression – price prediction, predict sales of product based on advertising budgets Forecasting - Predicting future sales based on past sales, Predictive maintenance, Weather forecasting Anomaly detection - Identifying transactions that are potentially fraudulent., Learning patterns that indicate that a network intrusion has occurred., Finding abnormal clusters of patients., Checking values entered into a system.
- https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetHighLevelConcepts.md#mlcontext
- IDV = Flexible, efficient way of describing tabular data The IDataView component provides a very efficient, compositional processing of tabular data (columns and rows) especialy made for machine learning and advanced analytics applications. It is designed to efficiently handle high dimensional data and large data sets. It is also suitable for single node processing of data partitions belonging to larger distributed data sets. Immutable – can’t change it – have to create a copy – not directly accessing the idv and changing it – making new copy of it Lazy – as needed go through it, not all in memory https://github.com/dotnet/machinelearning/blob/master/docs/code/IDataViewDesignPrinciples.md
- FlowerType is what we’re trying to predict. This becomes the Label property. This Label is the label or target variable. In an IDataView the column names are the names of the properties. The way in which data is loaded is determined by LoadColumn which specifies the index where to find that data point in the file. https://github.com/dotnet/machinelearning/blob/master/docs/code/IDataViewDesignPrinciples.md https://github.com/dotnet/machinelearning/blob/master/docs/code/SchemaComprehension.md
- https://docs.microsoft.com/en-us/dotnet/machine-learning/resources/transforms#feature-transformations https://docs.microsoft.com/en-us/dotnet/machine-learning/resources/transforms#feature-transformations Estimators and transformers: https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetHighLevelConcepts.md#transformer https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetHighLevelConcepts.md#estimator Estimators Untrained transformer Definition of the operations that are to take place Transformers Component that realizes the transformations defined by the estimators
- https://docs.microsoft.com/en-us/dotnet/machine-learning/resources/transforms#feature-transformations
- With ML.NET, the same algorithm can be applied to different tasks. For example, Stochastic Dual Coordinated Ascent can be used for Binary Classification, Multiclass Classification, and Regression. The difference is in how the output of the algorithm is interpreted to match the task. https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-choose-an-ml-net-algorithm
- https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetHighLevelConcepts.md#prediction-function
- https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/train-machine-learning-model-ml-net
- A statistical model or a machine learning algorithm is said to have underfitting when it cannot capture the underlying trend of the data. Its occurrence simply means that our model or the algorithm does not fit the data well enough.. The word overfitting refers to a model that models the training data too well. Instead of learning the genral distribution of the data, the model learns the expected output for every data point. This is the same a memorizing the answers to a maths quizz instead of knowing the formulas. Because of this, the model cannot generalize. Everything is all good as long as you are in familiar territory, but as soon as you step outside, you’re lost. Overfitting can be pretty discouraging because it raises your hopes just before brutally crushing them. https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/train-machine-learning-model-ml-net
- https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/train-machine-learning-model-cross-validation-ml-net https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetCookBook.md#how-do-i-train-using-cross-validation
- PFI measures feature importance by asking the question, “What would the effect on the model be if the values for a feature were set to a random value (permuted across the set of examples)?” FCC works by determining the amount each feature contributed to the model’s score for that particular data sample https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/explain-machine-learning-model-permutation-feature-importance-ml-net https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetCookBook.md#i-want-to-look-at-my-models-coefficients https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetCookBook.md#i-want-to-look-at-my-models-coefficients https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetCookBook.md#i-want-to-look-at-my-models-coefficients https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetCookBook.md#how-do-i-look-at-the-local-feature-importance-per-example
- Deep learning can be considered as a subset of machine learning. It is a field that is based on learning and improving on its own by examining computer algorithms. While machine learning uses simpler concepts, deep learning works with artificial neural networks, which are designed to imitate how humans think and learn. Until recently, neural networks were limited by computing power and thus were limited in complexity. However, advancements in Big Data analytics have permitted larger, sophisticated neural networks, allowing computers to observe, learn, and react to complex situations faster than humans. Deep learning has aided image classification, language translation, speech recognition. It can be used to solve any pattern recognition problem and without human intervention. Artificial neural networks, comprising many layers, drive deep learning. Deep Neural Networks (DNNs) are such types of networks where each layer can perform complex operations such as representation and abstraction that make sense of images, sound, and text.
- Low-level features are minor details of the image, like lines or dots, that can be picked up by, say, a convolutional filter (for really low-level things) or (for more abstract things like edges). High-level features are built on top of low-level features to detect objects and larger shapes in the image. Convolutional neural networks use both types of features: the first couple convolutional layers will learn filters for finding lines, dots, curves etc. while the later layers will learn to recognize common objects and shapes.
- Convolution is a general purpose filter effect for images In Convolutional Neural Networks, Filters detect spatial patterns such as edges in an image by detecting the changes in intensity values of the image. In terms of an image, a high-frequency image is the one where the intensity of the pixels changes by a large amount, whereas a low-frequency image is the one where the intensity is almost uniform. Usually, an image has both high and low frequency components. The high-frequency components correspond to the edges of an object because at the edges the rate of change of intensity of pixel values is high.
- Convolution is a simple mathematical operation which is fundamental to many common image processing operators. Convolution provides a way of `multiplying together' two arrays of numbers, generally of different sizes, but of the same dimensionality, to produce a third array of numbers of the same dimensionality.
- 3. Now when you apply a set of filters on top of that (pass it through the 2nd conv. layer), the output will be activations that represent higher-level features. Types of these features could be semicircles (a combination of a curve and straight edge) or squares (a combination of several straight edges). As you go through the network and go through more conv. layers, you get activation maps that represent more and more complex features.
- ML.NET uses TensorFlow through the low-level bindings provided by the TensorFlow.NET library. The TensorFlow.NET library is an open source and low-level API library that provides the .NET Standard bindings for TensorFlow. That library is part of the open source SciSharp stack libraries. To train image classification models, using the ML.NET API, use the ImageClassification API. You can also train custom deep learning models in Model Builder. The process is generally the same, but in addition to training locally, you can also leverage Azure to train models in GPU enabled compute instances.
- http://www.laurencemoroney.com/rock-paper-scissors-dataset/