H2O.ai Catalog
Extend the power of Driverless AI with custom recipes and build your own AI!
Extend the power of Driverless AI with custom recipes and build your own AI!
Parallel FB Prophet transformer is a time series transformer that predicts target using FBProphet models.This transformer fits one model for each time group column values and is significantly fasterthan the implementation available in parallel_prophet_forecast.py.
Data recipe to transform input audio to Mel spectrogramsThis data recipe makes the following steps:1. Reads audio file2. Converts audio file to the Mel spectrogram3. Save Mel spectrogram to .png image4. Upload image dataset to DAIRecipe is based on the Kaggle Freesound Audio Tagging 2019 challenge:https://www.kaggle.com/c/freesound-audio-tagging-2019To use the recipe follow the next steps:1. Download a subsample of the audio dataset from here:http://h2o-public-test-data.s3.amazonaws.com/bigdata/server/Image Data/freesound_audio.zip2. Unzip it and specify the path to the dataset in the DATA_DIR global variable3. Upload the dataset into Driverless AI using the Add Data Recipe optionThe transformed dataset is also available and could be directly uploaded to Driverless AI:http://h2o-public-test-data.s3.amazonaws.com/bigdata/server/Image Data/freesound_images.zip
Speech to text using Mozilla's DeepSpeechSettings for this recipe:Assing MODEL_PATH global variable prior to usageAssign WAV_COLNAME global variable with proper column name from your dataset.This colums should contain absolute paths to .wav file which needs to be converted to text.General requirements to .wav's:1 channel (mono)16 bit16000 frequency
Data recipe to transform input video to the images.This data recipe makes the following steps:1. Reads video file2. Samples N uniform frames from the video file3. Detects all faces on each frame4. Crops the faces and saves them as imagesRecipe is based on the Kaggle Deepfake Detection Challenge:https://www.kaggle.com/c/deepfake-detection-challengeTo use the recipe follow the next steps:1. Download a small subsample of the video dataset from here:http://h2o-public-test-data.s3.amazonaws.com/bigdata/server/Image Data/deepfake.zip2. Unzip it and specify the path to the dataset in the DATA_DIR global variable3. Upload the dataset into Driverless AI using the Add Data Recipe optionThe transformed dataset is also available and could be directly uploaded to Driverless AI:http://h2o-public-test-data.s3.amazonaws.com/bigdata/server/Image Data/deepfake_frames.zip
Speech to text using Azure Cognitive ServicesSettings for this recipe:Assing AZURE_SERVICE_KEY and AZURE_SERVICE_REGION global variable prior to usageAssign WAV_COLNAME global variable with proper column name from your dataset.This colums should contain absolute paths to .wav file which needs to be converted to text.