User Account Management
1. Please click Sign up to create a new account.
2. Afterwards, one can click Log in to log in to that account.
3. If you want to update your password, please select the "User Account" item on the main navigation bar, and follow the instructions to update the password.
Create and Evaluate a Predictor
1. After logging on: On the "Predictors" drop-down menu in the Navigation Bar, click "Create Predictor", then follow the directions: Give a name for this predict, and add some notes… the upload your training dataset (Click here for File Format). PSSP assumes your dataset contains between 20 to 30,000 instances.
2. If you wish to see an estimate of your predictor's performance (in terms of several measures), please check "Run Cross-Validation" box. (More information about training options can be found here).
3. Click "Submit". The training of your predictor can take anywhere from a few minutes to several hours depending on the size of your data set, and how busy the system is. The page will automatically reload while your predictor is being learned.
4. After the training is complete: If you specified Cross-validation, you can now examine the predictions for each instance, as well as various statistics. For example, for the NACD predictor [http://pssp.srv.ualberta.ca/predictors/32], these statistics appear in the main page [http://pssp.srv.ualberta.ca/predictors/32]; the individual predictions are in [http://pssp.srv.ualberta.ca/predictors/32/predictions]. Note you can sort these by the various fields, and click the [Details] button to see an individual prediction; for details see [URL].
Retrain an existing Predictor
1. To retrain an existing predictor (call it [predictor_id], of http://pssp.srv.ualberta.ca/predictors/[predictor_id] -- such as “32” above) , with different features or settings, open the "Prediction Setting/Retrain" section on the main page of a predictor ( http://pssp.srv.ualberta.ca/predictors/[predictor_id]) and click the "Retrain with Selected Options" button to navigate to the retrain page.
2. On the top of that retrain page, please select the features you want to use in the retrained version.
3. Please scroll down to the bottom of the retrain page and click the "Show Advanced Options" button to set the retrain options; see [URL].
4. Click the "Retrain Predictor" button at the bottom of the retrain page to start the retraining process.
Make Predictions using an Existing Predictor
1. After logging on: On the "Predictors" drop-down menu in the Navigation Bar, click "Use My Predictor" to navigate to prediction (analysis) page.
2. Select the predictor and upload your query dataset (Click here for File Format). The format of the data uploaded should be consistent with predictor's dataset.
3. Click "Submit". The prediction process should only take a few minutes. The page will automatically reload until your predictions are finished.
Feature Expansion
The survival dataset includes a set of instance, each described using a set of features. There are 3 different types of features:
1. The feature is considered a numerical feature, if all of the values could be parsed to numbers -- E. g., 2, 3.71, -19.90, etc.
2. The feature is considered a binary feature, if it only takes one of two values, and one of the values is a non-numerical attribute. E. g. {A, B}, {A, 1}, {A, blank}. Please note that {-1, 3} is considered a numerical feature rather than binary feature.
3. The feature is considered as a nominal feature, if it takes more than 2 values, and one of the values is not numerical -- E. g. {A, B, C}, {A, 1, 3}, {A, blank, 1}. In this case, each value in original feature corresponds to a new binary feature -- E. g., {A, B, C} become 3 features: A mapping to [1,0,0], B to [0,1,0] and C to [0,0,1].