Aligning the shape of tensors (or vectors/matrices) in a network can be challenging at times. As the network grows, it is common to lose track of dimensionalities in a complex expression. Instead of explicitly printing tensor shapes to debug, use 𝐓𝐞𝐧𝐬𝐨𝐫𝐒𝐞𝐧𝐬𝐨𝐫. It generates an elegant visualization for each statement executed within its block. This makes dimensionality tracking effortless and quick.
Deep Learning Network Debugging Made Easy
Deep Learning Network Debugging Made Easy
Deep Learning Network Debugging Made Easy
Aligning the shape of tensors (or vectors/matrices) in a network can be challenging at times. As the network grows, it is common to lose track of dimensionalities in a complex expression. Instead of explicitly printing tensor shapes to debug, use 𝐓𝐞𝐧𝐬𝐨𝐫𝐒𝐞𝐧𝐬𝐨𝐫. It generates an elegant visualization for each statement executed within its block. This makes dimensionality tracking effortless and quick.