upstride-alchemy 1.0.0a2

No project description provided

Alchemy CLI

this is the software development kit (SDK) and command line interface (CLI) for alchemy

install

to install this package run

pip install upstride-alchemy

or clone this repository and run

pip install [-e] .

or

python setup.py {install, develop}

SDK

To import the sdk in your project, do a simple

import alchemy

you can now login to alchemy using your credentials and initialize the project :

alchemy.login('your@email', 'yourpassword')
alchemy.init(project_name='project42', 
             run_name='mobilenet on cifar10', 
             dataset='cifar10', 
             model='mobilenet_v3', 
             tags=['pytorch'])

now you can simply call the log function to send results to the platform

alchemy.log(epochs=1, metrics={'accuracy': 0.8, 'loss': 3.1415, 'whatever metric you want': 0.9})

Command line interface

this repository provides a script to update results from a tensorboard checkpoint : alchemy_cli

One parameter is mandatory : log_file, the path of the tensorboard log file to parse.

The other parameter can be passed using the command line or will be asked by the script to the user

These parameters are:

• --user: email to connect to alchemy
• --password: password to connect to alchemy
• --step: step between two points to upload
• --project: Alchemy project to update
• --run: Alchemy run to update
• --tags: tags associated with the run (only if new run)
• --dataset: dataset used in run (only if new run)
• --model: Neural Network model used in the run (only if new run)
• --scalar_plots: scalar graph to upload

Every parameter except log_file can be stored in a yaml file, and provided with the yaml_file. An example yaml file can be:

user: your@email
project: my_project
run: my_run
scalar_plots:
 - my_metric1
 - my_metric2
tags:
 - tag1
 - tag2
dataset: my_dataset
model: my_model

This script can be run with

alchemy_cli <log_fie> <parameters>

Upload_everything

If your experiments are stored in a way that every event files comes with a corresponding yaml, you can use a dedicated script to upload runs based on a wildcard file pattern

bash upload_everything.sh "*my_pattern*" "my_password"