Please enable JavaScript.
Coggle requires JavaScript to display documents.
Machine Learning with Graphs - Coggle Diagram
Machine Learning with Graphs
Other skills
Pytorch tensor
Libraries
NetworkX
creating, manipulation for graphs and networks
analyze structure
Apply network algorithms
network / graph analysis
centrality measures
closeness
how close to all other notes
shortest path
degree
betweenness centrality
number of times node is on shortest path between other nodes
eigenvector
centrality of nodes the node is connected to
individual nodes
community detection
clusters or groups in a network
can detect communities
high centrality can help identify communities
focused on groups / clusters of nodes
network flow
optimal way to route stuff
maximizing flow
individual nodes
graph traversal
depth-first search
topological sort
Heterogenous Graphs
Adding node attributes
nx.classes.function.set_node_attributes
Torch (PyTorch)
Libraries / Extensions
PyTorch Geometric
make implementing and working with graph data more efficient
features
data handling
manipulate graph
loading graph datasets
layers
pre-implemented layers
GCN
GAT
GraphSage
Batch Processing
Message Passing
Layer Utilites
works with sparse data (like those in graphs)
extension library for GNN's
General Purpose ML library for creating and training Deep Learning Neural Networks
features
Traditional ML
Regression
Classification
NLP (Natural Language Processing)
Tensors | Multiple Dimensional Matricies for GPU
Optimization
torch,optim
update weights with respect to loss function
includes sophisticated loss methods
Vision
autograd | gradient on tensors
flexible input
variable input sizes
dynamic graph architecture
batch processing
data loader | custom data loading
GPU Acceleration
Dynamic Computation Graph
define by run (on the fly)
shapes, values, gradients
Scikit-learn
ML Algorithms
Supervised learning (classification)
Logistic Regression
Linear Regression
Clustering
Unsupervised Learning
Pre-processing and feature selection
model evaluation and tuning
Used for ML and Deep learning
Data needs to be clean and structured (like by Pandas)
Pandas
Data Structures
Series
Data Frame
Data manipulation
Data Analysis
Statistical Analysis
Time Series Analysis
File I/O
transforms formats
JSON
SQL
CSV
Excel
Handle missing data
Used for Prep: Clean, load, pre-process data
likes tabular data
DeepSnap
mew
Core Concepts
Node Degree (networkx)
Clustering
Average clustering coefficient (networkx)
Creating a Graph (Networkx)
Centrality
Closeness centrality (networkx)
Edges (networkx)
Embedding Layers
torch.nn.Embedding
number of embeddings
embedding dimensions (features)
data stored in a # of embeddings x # of dimensions matrix
SCD
torch.optim.SCD
Sigmoid
torch.nn.Sigmoid
maps any number to a value between 1 and 0
introduces non-linearity into a neural network model
Heterogenous Graphs
NetworkX
can set node attributes
manually by adding node labels and types as attributes
can set node edge attributes
manually