How do you experiment with machine learning approaches?

由人工智能和领英社区提供技术支持

Machine learning is a powerful and versatile field of computer science that involves creating systems that can learn from data and perform tasks such as classification, regression, clustering, anomaly detection, and more. However, machine learning is not a one-size-fits-all solution, and different approaches may have different advantages and disadvantages depending on the problem, the data, the goals, and the constraints. Therefore, experimenting with machine learning approaches is a crucial skill for any aspiring or practicing machine learning engineer or researcher. In this article, we will discuss some general steps and tips for conducting effective and efficient experiments with machine learning approaches.

此文章中的业界达人

由社区从 7 条内容中精选。了解更多

Pooja Palod

Data and Applied Scientist ll at Microsoft |Building Machine Learning Systems |IIT Bombay | AIR 51, Gate CSE 2016

1 Define the problem and the metrics

The first step in any machine learning experiment is to clearly define the problem you want to solve and the metrics you want to use to measure the performance of your approach. The problem definition should include the input and output variables, the type of task (e.g., classification, regression, etc.), the domain and scope of the problem, and any assumptions or constraints. The metrics should be aligned with the problem definition and reflect the desired outcomes and trade-offs. For example, if you are working on a spam detection problem, you may want to use precision and recall as your metrics, and balance them according to your preference for false positives and false negatives.

添加您的观点

Pooja Palod

Data and Applied Scientist ll at Microsoft |Building Machine Learning Systems |IIT Bombay | AIR 51, Gate CSE 2016
(已编辑)
举报内容
Its very important to define scope of the ml project you will be working on. To start with you need to have clear understanding of business problem you are trying to solve.This can be done by collaborating with domain experts and business stakeholders or in same case product mangers to understand challenges and goals.Once this is done you need to articulate it into machine learning problem statment clearly defining what you want to achieve and how you will measure it(using metrics).You also need to decide on type of problem (regression,classification,ranking etc),features and target variable.

已翻译

赞
Md Azimul Haque

Engineering Manager | Machine Learning | Data Science
举报内容
We should choose statistical metrics based on how close they are to business metrics. For example, if you want to predict gas prices accurately, the metric can be mean absolute error (MAE). We should also consider data attributes for selecting metric. For example, if the dependent variable has class imbalance, we need to select F1 score, as it balances both precision and recall.

已翻译

赞

2 Explore the data and preprocess it

The next step is to explore the data you have available for your problem and preprocess it to make it suitable for your machine learning approach. Data exploration involves analyzing the distribution, quality, and characteristics of the data, such as its size, shape, type, range, outliers, missing values, noise, and correlations. Data preprocessing involves transforming the data to make it more compatible with your machine learning approach, such as scaling, encoding, imputing, cleaning, filtering, augmenting, and splitting. Data exploration and preprocessing can help you gain insights into your problem, identify potential issues or opportunities, and improve the performance and robustness of your machine learning approach.

添加您的观点

Md Azimul Haque

Engineering Manager | Machine Learning | Data Science
举报内容
Data exploration can be done in 2 parts. In the first part, we should try to identify data quality issues and try to fix it. Some of these problems can be outliers, missing data, and any other detected pattern which counters business logic. The second part should be about exploring and identifying interaction effects amongst features.

已翻译

赞

3 Choose a machine learning approach and implement it

The third step is to choose a machine learning approach that matches your problem and your data, and implement it using a programming language or a framework of your choice. A machine learning approach consists of a model, which is the structure or architecture of the system that learns from the data, and an algorithm, which is the process or method that trains the model using the data. There are many types of models and algorithms available for different machine learning tasks, such as linear models, decision trees, neural networks, support vector machines, k-means, etc. You can also customize or combine different models and algorithms to suit your needs. To implement your machine learning approach, you can use a low-level language like Python or C++, or a high-level framework like TensorFlow or PyTorch, depending on your preference and expertise.

添加您的观点

Md Azimul Haque

Engineering Manager | Machine Learning | Data Science
举报内容
We can develop 2 baseline models for machine learning problems. The first can be a linear model and the second is based on non-linear model. For linear model, we should check if model assumptions are getting violated and try to fix it. For non-linear model, we should try feature selection, hyperparameter turning etc to find an optimal model. Finally, the model which performs the best should be selected.

已翻译

赞

4 Train and evaluate your machine learning approach

The fourth step is to train and evaluate your machine learning approach using the data you have preprocessed and split into training, validation, and test sets. Training involves feeding the training data to your machine learning approach and adjusting its parameters to minimize a loss function that measures the error between the predicted and actual outputs. Evaluation involves testing the performance of your machine learning approach on the validation and test data using the metrics you have defined, and comparing it with a baseline or a benchmark. Training and evaluation can help you assess the quality and generalization of your machine learning approach, and identify any problems or limitations, such as underfitting, overfitting, bias, variance, etc.

添加您的观点

Md Azimul Haque

Engineering Manager | Machine Learning | Data Science
举报内容
Once the model is performing well on training and validation data, we can finally check the performance on external test data. If the model is performing well on all the datasets in a consistent manner, we can consider that model will generalize well on unseen data.

已翻译

赞

5 Tune and optimize your machine learning approach

The fifth step is to tune and optimize your machine learning approach to improve its performance and efficiency. Tuning involves adjusting the hyperparameters of your machine learning approach, which are the settings or options that control its behavior and complexity, such as the learning rate, the number of layers, the regularization factor, etc. Optimization involves applying techniques or methods that enhance the speed, memory, or accuracy of your machine learning approach, such as gradient descent, stochastic gradient descent, batch normalization, dropout, etc. Tuning and optimization can help you find the optimal configuration and solution for your machine learning approach, and balance the trade-offs between performance and efficiency.

添加您的观点

Md Azimul Haque

Engineering Manager | Machine Learning | Data Science
举报内容
There are multiple approaches for hyperparameter tuning. Grid search is one such popular method. We can also use the Optuna python library for hyperparameter optimization.

已翻译

赞

6 Experiment and iterate

The final step is to experiment and iterate with different machine learning approaches, data sets, metrics, and techniques, and analyze the results and feedback. Experimenting involves trying out different combinations and variations of the previous steps, and observing how they affect the outcome and behavior of your machine learning approach. Iterating involves repeating and refining the previous steps based on the results and feedback, and incorporating new ideas or improvements. Experimenting and iterating can help you discover new insights, patterns, or solutions for your problem, and learn from your successes and failures.

添加您的观点

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?

添加您的观点

Pooja Palod

Data and Applied Scientist ll at Microsoft |Building Machine Learning Systems |IIT Bombay | AIR 51, Gate CSE 2016
举报内容
Model metric you use should align with your business objective.For example lets say your are building model for customer churn prediction then your model metric could be recall(find all customers you have high probability of closing the subscription) .This will perfecty align with business metric of customer churn rate.

已翻译

赞

Computer Science

+ 关注

给文章评分

我们借助人工智能创建了此文章。您认为这篇文章怎么样？

很棒不太好

举报此文章

查看全部

How do you experiment with machine learning approaches?

1

2

3

4

5

6

7

1 Define the problem and the metrics

2 Explore the data and preprocess it

3 Choose a machine learning approach and implement it

4 Train and evaluate your machine learning approach

5 Tune and optimize your machine learning approach

6 Experiment and iterate

7 Here’s what else to consider

Computer Science

给文章评分

感谢您的反馈

更多Computer Science相关文章

更多相关阅读内容

How do you experiment with machine learning approaches?

1

2

3

4

5

6

7

1 Define the problem and the metrics

2 Explore the data and preprocess it

3 Choose a machine learning approach and implement it

4 Train and evaluate your machine learning approach

5 Tune and optimize your machine learning approach

6 Experiment and iterate

7 Here’s what else to consider

Computer Science

给文章评分

感谢您的反馈

查看其他技能