{"id":6544,"date":"2021-01-19T16:34:32","date_gmt":"2021-01-19T16:34:32","guid":{"rendered":"https:\/\/www.blopig.com\/blog\/?p=6544"},"modified":"2021-01-19T16:34:38","modified_gmt":"2021-01-19T16:34:38","slug":"tracking-machine-learning-projects-with-weights-biases","status":"publish","type":"post","link":"https:\/\/www.blopig.com\/blog\/2021\/01\/tracking-machine-learning-projects-with-weights-biases\/","title":{"rendered":"Tracking machine learning projects with Weights & Biases"},"content":{"rendered":"\n

Optimising machine learning models requires extensive comparison of architectures and hyperparameter combinations. There are many frameworks that make logging and visualising performance metrics across model runs easier. I recently started using Weights & Biases. In the following, I give a brief overview over some basic code snippets for your machine learning python code to get started with this tool.<\/p>\n\n\n\n\n\n\n\n

For account setup see here<\/a>. The basic account is free.<\/p>\n\n\n\n

First, initialise the run alongside all parameters you want to keep track off before you start the model training:<\/p>\n\n\n\n

wandb.init(\n    project='demo_project',\n    name='demo_run',\n    config={\n        'learning_rate': 0.001,\n        'dropout': 0.2,\n        'layers': 6,\n        'train epochs': 3\n    }\n)<\/pre>\n\n\n\n
 This will set up a project ‘demo_project’ in your Weights & Biases account and log the following code as the run ‘demo_run’, associating the hyperparameters logged in config with the run metrics.<\/p>\n\n\n\n
We can then log any metrics to this run as demonstrated here:<\/p>\n\n\n\n
for epoch in range(1, 5):\n    for step in range(1, 100):\n        \n        ####\n        # Model training goes here\n        ####\n\n        # simulating loss and accuracy of train step\n        dummy_loss = mult * (1\/(step + 5 * epoch) + random.random()\/(step + 5 * epoch))\n        dummy_accuracy =  1 - mult * (1\/math.sqrt((step + 5 * epoch) + random.random()))\n        \n        # log to Weights & Biases\n        wandb.log({'Train loss': dummy_loss})\n        wandb.log({'Train accuracy': dummy_accuracy})\n\n    ####\n    # Model testing per epoch goes here\n    ####\n    \n    dummy_test_loss = mult * (1 \/ (epoch * 10) + random.random() \/ (epoch * 10))\n    dummy_test_accuracy = 1 - mult * (1\/math.sqrt(epoch * 10 + random.random()))\n    wandb.log({'Test loss': dummy_test_loss})\n    wandb.log({'Test accuracy': dummy_test_accuracy})<\/pre>\n\n\n\n
This can the be used for easy visualisation of training and testing progress:<\/p>\n\n\n\n