Train
Last updated
Was this helpful?
Last updated
Was this helpful?
In the training section, you can view general information about the model, including the number of training and test data
The "Train" section of the website consists of three main subsections: "PROJECT SPEC," "Train Model," and "Train Result."
PROJECT SPEC: In this section, the details and specifications of the project for training machine learning models are outlined. This includes goals, requirements, constraints, and other details related to the training process. This document serves as a general guide for the training team and other team members involved in the training process.
Train Model: This section describes the process of training machine learning models using the dataset prepared earlier. It includes selecting and configuring training algorithms, conducting training, and the steps involved in evaluating and optimizing the models.
Train Result: In this section, the results and progress obtained during the training process of machine learning models are reported. This includes evaluation metrics, training results, charts, and any other type of output generated during the model training process.
In the "PROJECT SPEC" section, you can explore information regarding the dataset, details, data augmentation, sample images, and the "dataset healthcheck" option.
Overall, the "Train" section is used for training machine learning datasets and provides details about the process of training models.
In the "TRAIN MODEL" section, you can utilize the framework, model, and other details for training your model.
You can use the following filters as needed and depending on your goals:
Deployable on GPU
Deployment on cloud
High accuracy
Fast results
Deployable on CPU
Optimization
Certainly, here's an explanation for each of these items:
Deployable on GPU: This refers to the capability of a model or application to be deployed and run on Graphics Processing Units (GPUs). GPUs are specialized hardware designed for parallel processing and are often used for accelerating deep learning inference tasks due to their high computational power.
Deployment on Cloud: This indicates the ability to deploy a model or application on cloud computing platforms such as Amazon Web Services (AWS), Microsoft Azure, or Google Cloud Platform (GCP). Deploying on the cloud offers scalability, flexibility, and accessibility advantages, allowing users to access resources remotely without the need for on-premises infrastructure.
High Accuracy: High accuracy suggests that the model or system achieves a high level of correctness or precision in its predictions or outcomes. In the context of machine learning models, accuracy is typically measured by comparing the model's predictions to ground truth labels and evaluating metrics such as precision, recall, F1-score, or classification accuracy.
Fast Results: This implies that the model or system can produce outputs or predictions quickly, with minimal latency or delay. Fast results are crucial in real-time or time-sensitive applications where rapid decision-making is required, such as autonomous driving, real-time object detection, or video processing.
Deployable on CPU: Similar to "Deployable on GPU," this refers to the ability of a model or application to be deployed and run on Central Processing Units (CPUs). CPUs are general-purpose processors commonly found in most computing devices and are suitable for running a wide range of software applications, including machine learning models that do not require the computational power of GPUs.
Optimization: Optimization involves the process of improving the performance, efficiency, or resource utilization of a system or model. In the context of machine learning, optimization may include techniques such as model compression, quantization, pruning, or algorithmic improvements to reduce memory footprint, inference latency, or energy consumption while maintaining or improving accuracy.
In the "TRAIN DETAILS" section, you can view the status of the models you have requested for training. As depicted in the image below, the models submitted for training have three statuses:
Queue
Done
Failed
As depicted in the image, in box 1, you can observe filters for searching trained models. In box 2, you can see the status of the models. In box 3, you can find metrics of the trained model such as MAP and recall. If the model fails to pass the health check, the progress status or the reason for failure is specified, similar to what you see in box 4.
Additionally, in box 5, you can view details and logs, archive the model, or access the console.
By clicking on the trained model, you can:
View details including:
Epochs
Batches
Max Size
Patience
Momentum
Learning Rate
Test the trained model and visualize its tensorboard.
In the "Model Train Result" section, you can easily view the model outputs on test and validation datasets.