Skip to content

TAZI Hub Sport Mode

You can change TAZI Hub to Sport mode by clicking Sport button. Optimize and ML Ops sections will appear after clicking it.

Also, Confidence and Benefits tabs will be added to the Observe section as you can see below.

Optimize

Human in the Loop

This module allows domain users to provide instance or model level feedback to teach the machine learning system. This results in models learning faster while data changes. When the domain user feedback is provided, the machine learning system may need labels only for a portion of the data. This means less expert time labeling and less time waiting for the results to accumulate. Click here to view Human in the Loop in detail.

Human in the loop can be used while the Business Model is being run. Click on Human in the Loop to start giving instant feedback to the Business Model:

image-20210725215136391

Model Based Feedback

image-20210725214635594


The user interacts with a continuously learning model through its explanation model and may change the assigned label of a segment, Tazi internally feeds these relabeled samples to the learning path to move the model in the desired direction.

Instance Based Feedback

image-20210725214740047

While training Tazi puts aside some samples to present on demand for user feedback; the sample selection is made such that it would improve the model in the areas it is less certain. The user may change or confirm Tazi's decisions for these instances, then they are fed back into Tazi's learning path.

Create HYPO

By clicking this button, a new HYPO Model will be created using the data source and configuration of the selected Business Model. Thus, you don't need to fill in all those details again. You can quickly create new HYPO Model by just configuring HYPO options and parameters.

You can create a Hyper-Parameter Optimization (Hypo) model by clicking Create Hypo button below:

image-20210725132411375


It provides to create new HYPO Model by using the data source and configuration of the selected business model. So you don't need to fill in all those details again. You can quickly create new HYPO Model by just configuring HYPO options and parameters:


image-20210725134717392


In HYPO Options tab, you can find general settings of the Hypo configuration that will be used to optimize specified parameters’ values to get better performance for the run.

  • Population Size : It sets first random population size. This means, there will be 4 candidates (chromosomes) in first iteration.
  • Iteration Count : It sets number of evolution of current population. Iteration Count must be bigger than 1. Minimum value should be 2 (two) to calculate fitness values of random initialized population.
  • Number of Simultaneous Runs : It sets how many instances can run at same time (parallel). Since creating parallel tasks may cause performance issues, it is recommended to keep in between 3 (three) and 5 (five).

When you click Advanced Settings switch, a new set of settings will appear:

  • Selection Method : Best, Tournament, Threshold and Roulette Wheel methods are available.
  • Crossover Method : New chromosomes are generated with crossover. Currently available methods:
  • One Point: A random point selected over two parent chromosomes, first part of parent1 combined with second part of parent2, vice versa.
  • Two Point: Random two points selected over two parent chromosomes, middle part of parent1 inserted two ends of with second part of parent2, vice versa.
  • Mutation Rate : Mutation rate of JGAP is 1/12. We are able to change this behavior by using mutationRate configuration. It should be between 1 (one) and 12 (twelve). If you set it to 1 (one), this means every genes except MultiGene in chromosome will be mutated (this also means mutation rate is nearly 70% because MultiGenes can not be mutated). If you want high diversity, set it to 1 (one). You can set it to 6 (six) which makes mutation rate nearly 30%. It is recommended to use it with 1 (one) or 6 (six).
  • Model Metric : Fitness values of each run can be set to one of TAZI’s performance metrics:
  • Accuracy
  • Weighted Accuracy
  • Equal Weighted Accuracy
  • F1 Score
  • Recall
  • Precision
  • Top K
  • Auto Termination : This feature provides to stop Hypo Run automatically. Default value is false.


When you are done with the settings you can decide on which paremeters to optimize by clicking Parameters to Optimize tab:


image-20210725140027428

Click on Add Parameter and start choosing your parameters:


image-20210725140315310


All of the parameters related your Business Model configuration will be found here. You can select one and assign possible values the parameter can take depending on its type:


image-20210725140555713


Here, we have chosen Alpha Confidence as our parameter to optimize. 0.95 value is already added since it was its default value when we configured our Business Model. We add extra values to so Hypo model can run our Business Model with different values of this parameter and decide which one yields a better result in terms of a selected performance metric.

After clicking Done we can see our newly added parameter in the parameter list:


image-20210725141955837

We can edit and/or delete the parameter.

We can choose to load all of the parameters from our Business Model configuration and edit or delete the ones we don’t need:

image-20210725142538218

Now, let’s switch to the Parameters to Copy tab:


image-20210725143748330


Parameter values can be copied from other parameters with determined coefficient. This is useful for the labelConfusionCosts parameters. One of them can be used as a gene, others would copy the first ones' value.

  • Name : Parameter will be updated according to the given parameter in the ‘Copy Value From’ section.
  • From : Select parameter’s value will be copied.
  • Coefficient : Multipler value. It can of type Int or Double

Now that we are done with the settings, let’s head over to the left side of the window:


image-20210725161926861


Before clicking Next, do notice that you can copy hypo configuration same as you can with Business Models. If there are hypo models created before their configurations will be listed here. You can simply click the check icon next to configuration to copy all of the settings needed to set up a hypo model. We can now click Next:


image-20210725162324385


Now, let’s type in a name and a description for our hypo model and click Submit. You’ll see the hypo model that we have just created in the Hypo Models list below:


image-20210725164702655


All of the created and completed hypo models will be listed here.

#: Unique hypo model number that is set automatically after the model creation.

Name: Name of your model.

Group: Group allows the TAZI administrator to configure authorization settings for users with respect to groups they belong to. If group is selected as Private, the hypo model will only be accessible to the TAZI administrator and the creator of the hypo Model. If a certain group name is selected, the access to the hypo model is restricted to the members of the selected group only.

Status: One of the following can be seen at Status column: CREATED, INITIALIZING, RUNNING, COMPLETED, STOPPED.

Progress: This shows how many iterations are left out of the total iteration count selected while configuring the hypo model.

Current Best: It show the best individual result according to the configured Hypo parameter set.

Action: When you click the icon in Action column, you’ll see a list of available actions:

  • Start: Starts the hypo model run.

  • Configure Hypo Parameters: You can configure hypo model parameters.

  • Configure Model Parameters: This will let you configure the underlying Business Model’s parameters the hypo model is based on.

  • Edit: You can edit name, description and group information of hypo model.

  • Stop: This stops the hypo model run.

  • Get Logs: You can see logs of outputs and errors regarding the run of hypo model.

  • Delete: You can delete the hypo model by clicking this.

  • Go to Hypo Model Details: After clicking on this, you will see a screen like below. On the top right section you can see which run has best result. In our example it is iteration:2 run:1 one as you can see.

image-20210725171215857

At the lower part of the screen you see Hypo Iteration Results. In this section you can see detailed results for each run by expanding iterations and runs.

image-20210725171335810

You can click Show Parameters button to see which hyperparameter set is responsible for this result. When you click Export Parameters button, it will download a txt file including hyperparameter values. Also, you can start a new run with these settings by clicking Create New Run button.

Advanced Parameters

In the Advanced section, you can tune more advanced settings of the way TAZI works and runs your model.

Go to Advanced section for detailed information.

ML Ops

Concept Drift

will be updated

Data Drift

will be updated