Orange: Nomogram

Sumber: https://docs.biolab.si//3/visual-programming/widgets/visualize/nomogram.html

Widget Nomogram untuk mem-visualisasi classifier Naive Bayes dan classifier Logistic Regression.

Input

Classifier: trained classifier
Data: input dataset

Output

Features: selected variables, 10 by default

Widget Nomogram memungkinkan representasi visual beberapa classifier (tepatnya Naive Bayes classifier dan Logistic Regression). Widget Nomogram menawarkan wawasan tentang struktur data training dan efek atribut pada probabilitas class. Selain visualisasi dari classifier, widget Nomogram menawarkan dukungan interaktif untuk prediksi probabilitas class. Cuplikan di bawah ini menunjukkan nomogram dari dataset Titanic, yang memodelkan probabilitas dari penumpang yang tidak survive dalam kecelakaan Titanic.

Pada saat banyak atribut yang bisa di plot dari dataset, hanya yang rank terbaik yang akan di pilih untuk di tayangkan. Kita mungkin memilih dari ‘No sorting’, ‘Name’, ‘Absolute importance’, ‘Positive influence’ dan ‘Negative influence’ untuk representasi Naive Bayes dan dari ‘No sorting’, ‘Name’ dan ‘Absolute importance’ untuk representasi Logistic Regression.

Probabilitas dari target class yang dipilih di hitung dengan prinsip ‘1-vs-all’ , yang sebaiknya di ambil sebagai prinsip dalam memilih multiclass data (jika total probabilitas tidak 1). Untuk menghindari ketidak nyamanan ini, kita dapat me-normalisasi probabilitas.

• Select the target class you want to model the probability for. Select, whether you want to normalize the probabilities or not.
• By default Scale is set to Log odds ration. For easier understanding and interpretation option Point scale can be used. The unit is obtained by re-scaling the log odds so that the maximal absolute log odds ratio in the nomogram represents 100 points.
• Display all attributes or only the best ranked ones. Sort them and set the projection type.

Continuous attributes can be plotted in 2D (only for Logistic Regression).

Contoh

Widget Nomogram widget sebaiknya digunakan langsung sesudah training widget classifier (yaitu, Naive Bayes atau Logistics Regression). Widget Nomorgram juga dapat mengirimkan data instance menggunakan widget yang memungkinkan selection (seperti widget Data Table) seperti tampak pada workflow di bawah ini.

Referring to the Titanic dataset once again, 1490 (68%) passengers on Titanic out of 2201 died. To make a prediction, the contribution of each attribute is measured as a point score and the individual point scores are summed to determine the probability. When the value of the attribute is unknown, its contribution is 0 points. Therefore, not knowing anything about the passenger, the total point score is 0 and the corresponding probability equals the unconditional prior. The nomogram in the example shows the case when we know that the passenger is a male adult from the first class. The points sum to -0.36, with a corresponding probability of not surviving of about 53%.

Features output

The second example shows how to use the Features output. Let us use heart_disease data for this exercise and load it in the File widget. Now connect File to Naive Bayes (or Logistic Regression) and add Nomogram to Naive Bayes. Finally, connect File to Select Columns.

Select Columns selects a subset of variables, while Nomogram shows the top scoring variables for the trained classifier. To filter the data by the variables selected in the Nomogram, connect Nomogram to Select Columns as shown below. Nomogram will pass a list of selected variables to Select Columns, which will retain only the variables from the list. For this to work, you have to press Use input features in Select Columns (or tick it to always apply it).

We have selected the top 5 variables in Nomogram and used Select Columns to retain only those variables.

Referensi

• https://docs.biolab.si//3/visual-programming/widgets/visualize/nomogram.html

Pranala Menarik

• Orange