TorchUtils is a Python package providing helpful utility APIs for your PyTorch projects.

Features:

• Save/load checkpoints.
• Calculate dataset statistics (mean, std, var). Also calculate and track running statistics of data.
• Get/set learning rate.
• Track evaluation metrics such as accuracy, running loss, hamming loss.
• Print model summary. Supports: Linear/MLP, Convolution Network, Recurrent Network (RNN/LSTM/GRU), Recursive Network.
• Calculate model FLOPs.
• Calculate total model parameters.
• Set random seed.
• Visualize gradient flow in your network.

Requirements:

• PyTorch >= 1.0.0
• Numpy >= 1.16.2
• Matplotlib >= 3.0.3