Ducousso, Mme Soline (2019) Probabilistic type inference PRE - Research Project, ENSTA.
![]()
| PDF 786Kb |
Abstract
The type inference problem is undecidable, that is to say not suitable for deterministic algorithms. Then the type inference problem has been apprached with probabilistic type inference and machine learning. Previously used models are sequential and do not take into account the structure of a source code, that is why we chose to implement a Gated Graph Neural Network model for python. This models requires a graph as input, computed form the Abstract Syntax Tree of Github [1] repositories. In this process, we gather type annotations which can be arbitrarily complex. We then build a type lattice to be able to generalize a rare type and design more precise metrics.
Item Type: | Thesis (PRE - Research Project) |
---|---|
Uncontrolled Keywords: | Python, Github, type annotations, machine learning, Gated Graph Neural Network, Abstract Syntax Tree, type lattice. |
Subjects: | Information and Communication Sciences and Technologies |
ID Code: | 7567 |
Deposited By: | Soline Ducousso |
Deposited On: | 18 juill. 2023 12:23 |
Dernière modification: | 18 juill. 2023 12:23 |
Repository Staff Only: item control page