ChronoScanML Tools 1.0.1-beta
User guide
×
  (c) ChronoScan Capture, 2023

Training

 
When we have annotated  our batch  and/ or have a dataset ready for training we enter this section.
 
These are the options:
 
Training
1

Train model

1. Train model    
Click here to start training.
*Before training, we have to create a valid dataset with the corresponding format depending on the architecture.
Check "Exporting a batch to a ML Dataset"
2

Training configuration

2. Training configuration
Edit the training configuration here.
You can edit some hyperparameters and other options for the training.
The information for each option is displayed when you click on its name.

 
3

Open training folder

3. Open training folder
It opens the training directory for the model.
Training dataset and saved training models are here.
4

Open model folder

4. Open model folder
Directory of the processing model. This is, the one that is "active" for inferences.
5

Refresh

5. Refresh
Refresh button for updating the trainings grid list.
6

Build info

6. Build info
Information prompt for the training executable.
 
The trainings grid list
 
Training
1

Activate model

1. Activate model
It activates a saved trained model, overwriting the current (if any) for inferences. It becomes the current processor model.
2

Delete model

2. Delete model
It deletes the trained and saved model
3

Open report

3. Open report
When a training attemp its successful and saved, a report is generated with relevant information about the training itself.
This is a set of stats and configurations used for the training.
This report is useful for sharing information and also to compare results with other trainings.
This is an example of the report:
 
 
4

Open folder

4. Open folder
It opens the saved model directory
5

Tensorboard

5. Tensorboard
TensorBoard is a web-based visualization tool provided by TensorFlow, a popular deep learning framework.
TensorBoard helps in understanding and debugging the training process, comparing different experiments, and gaining insights into the model's performance and behavior.
here's more information about it: https://www.tensorflow.org/tensorboard?hl=es-419
In order to be able to use TensorBoard , your machine must have python >=3.9 installed with the corresponding tensorboard dependecies.
     pip install tensorflow
     pip install tensorboard
When clicking this button we are starting TensorBoard on the port 6006 pointing to the lightning_logs located under the training directory.
a browser will open with the stats of your training:
* If the training just started it might take a while to start showing results and requires to be refreshed until so.
** If you encounter problems opening TensorBoard , you can also navigate (cmd) to the training folder of the model and run the following command with the desired port:
 
>path\Train_20230703133241>tensorboard --logdir=lightning_logs --port=PORT_NUMBER



 
 
When training, several models are being created in different checkpoints, but at the end of the training, only the best model and its corresponding tokenizer will be kept and saved.
 
The help manual was created with Dr.Explain