This repository is a compilation of scripts that I have created in my time working with machine learning. These scripts aim to automate the annoying and tedious parts of ML, allowing you to focus on what is important.\nPyPi: https://pypi.org/project/AutoAILib/\n $ pip install autoailib \nThis library was developed for and used with keras convolutional neural networks. They do however work with other keras models, besides image test obviously.
\n\n
AutoAi.convnet_tester(model path or object, labels)
\n
Class Video Demo \n
AutoAi.image_predict(model_path, image_path, labels)
\n
This function takes 3 arguments: a path to a keras model, a path to an image, and a list of labels.
\n
Demo:
\n Given a the correct arguments, we get the following output, as well as this image saved to our model directory.\n
![](\"https://i.imgur.com/woiPdus.png\")
\n
AutoAi.manual_test(model, testing_dir, labels)
\n
This function tests a model given labels and testing data. It then compiles the results in a CSV file, and groups the results by class, and by correct and incorrect.
\n
\n - Model - Path of model that you want to test or model object.
\n - Testing_dir - Path to the directory with your testing data.
\n - Labels - Dictionary of the classes, in form (index:class_name)
\n
\n
Example csv:
\n
![](\"https://i.imgur.com/g4gNQjS.png\")
\n
\n
Update! This has now been packaged in the AutoAI.data_compiler class.\n AutoAi.data_compiler(self,src, dest, **kwargs)\n AutoAi.data_compiler.run() will compile the data based on the constructor parameters.
\n
This function takes 2 required arguments, an original data source file, and a path to the desired data directory. Given just these two arguments, this function will create a new testing data folder at dest with training, validation, and testing folders, containing folders for each class. You can alter the ratio with the ratio arguments, as well as provide a number of img transforms to do if you are using images.
\n
\n - Src - Path to a folder that contains a folder for each class and then data examples in those class folders.
\n - Dest - Path to a folder where you want the data to end up.
\n - Num_imgs_per_class - This number of images will be added to the original set for each class through transforms. The theoretical limit for this would be 3! * original images per class
\n
\n
Demo:
\n Given the a path to the following folder:\n
![](\"https://i.imgur.com/SSpydEv.png\")
\n If augmentation used the following results will be yielded:\n
![](\"https://i.imgur.com/4okyMrN.png\")
\n Then these images will be copied to the dest folder with copied file structure, but an added upper layer:\n
![](\"https://i.imgur.com/TY7HvL4.png\"</img)
\n Example showing the images made it:\n
![](\"https://i.imgur.com/3ily5dU.png\"</img)
\n