When to use transpile(), assemble(), execute() and run() in qiskit?

What's the difference between them?

Yeah, this is very simple to understand. First let me get through the difference between run() vs execute() - unlike execute() , run() require the circuit to be transpiled first. The compile() refers to conversion of high level language to low level language where as transpile() is all about conversion of a high level language to another high level language. Assemble() assembles the circuit into a Qobj that can be run. Most generally, assemble() is used when working with Aer, where as transpile() used when working with IBMQ Provider.

qobj = assemble(qc)  # Assemble circuit into a Qobj that can be run
counts = sim.run(qobj).result().get_counts()  # Do the simulation, returning the state vector
plot_histogram(counts)  # Display the output on measurement of state vector