If you are like me, you are fascinated with Artificial Intelligence and Machine learning. But beyond fascination there are already practical uses. Imagine that your bubble application could use image and speech recognition or feed the AI dragon some energy (data) and produce incredible results.
Has anyone thought of, or is in the process of building an AI-Bubble plugin? For example to the Google CLOUD MACHINE LEARNING ENGINE or other AI APIs? Or some how integrating with Google TensorFlow?
I am willing to consider venturing with others into this area. As an freelancer I have to watch my time but my interest will be a driving force.
I could help too. I am actually working on the other end of the spectrum and already have trained models in Microsoft and Google. I just need help in figuring out how to connect it with Bubble and we are good to go.
I am just talking access the Google AI API. From what I understand, to train a data set, you need to use something like TensorFlow with something like Python.
I am going to try #1with a simple REST to the Google API and get back with you.
Dabbling with Amazonās platform it is fairly easy to train an algorithm by pushing your Bubble data to csv into Amazon S3. I have not yet used their API, but it also seems straight forward to call your algorithm with formatted user data to return the result. Charges were reasonable if you have a revenue generating app.
@mente12
Here is a suggestion: āLets aim for creating an image recognition bubble app that can detect patters?ā
What you do - Go to Microsoft Cognitive Services or Google Vision API and you already get access to trained set of data that can detect objects etc.
Challenge for me: I am more of a DB/Core Java/Design kinda guy, and their-in lies my strength and I have no idea how to call the APIās they use from Bubble. There is a API for credential and another one for uploading images and finally one for running the algorithm and getting the 5% correlation back.
If you can tell me how to do the above to Bubble (and we should collaborate on this) - I already have a trained data set on Microsoft which can detect patterns with >80% accuracy and we can test it out.
Hi,
Using the Blockspring plugin gives you a lot of options. Iām currently using the Monkey Learn integration for text classification and sentiment analysis.
Blockspring also has other integrations for image analysis (Clarafai if I remember correctly).
Also, I played around with Api.ai awhile ago for NLP, but it didnāt have a Blocksping integration yet and I couldnāt get it to play nicely with Bubbleās connector.
Jarrad,
If you want to keep it general and widely applicable, Feeding in a series of data points (with the first set containing the āanswerā) and training a model. Then feeding in new data (w/o the answer) and receiving a prediction.
An example would be predicting the temperature on a given day, using day of the year, yesterdayās temperature and location latitude. First, feed last yearās data into the model for training, and expose the resulting algorithm for use. This format should work for many problems.
I assume linear & logistic regression / decision trees / Naive Bayes would cover most peopleās use cases. Perhaps Cluster models would be nice as well.
mente - Iād consider using BigML to do the training outside of bubble, then utilize the API connector to make/retrieve your prediction. Alternatively, you can both make and retrieve the model via the API if your use case demanded it, but the above way is much quicker to implement.
PS: (There is also a connector through BlockSpring, but that does not appear to have the ability to make predictions)
Thanks for the suggestion. BTW: The other day I attempted to use BigML using the BlockSpring connector, however I believe for this to work, I need to be in a paying account at BigML.
I am guessing since I was getting an error although I have green checkmarks on all the authentication values.
Not sure about it being on the BigML side as I do not have a BigML account and was able to successfully use the API Connector to POST a prediction request for an Iris dataset that I created. I believe BigML is free for models up to 16MB in both prod and dev environments.
Might be a BlockSpring error you ran into instead.
Great thread - what different kids of AI software can you create within Bubbleās capabilities? Iām trying to advise my clients how far they can ādreamā without wasting their time.
Any examples of AI business models that Bubble can handle are greatly appreciated (and may lead to future hiring opportunities of either you as a developer or your template). Thanks!
You can make calls to existing AI APIs. For example pass a string of characters and let the Google pre-trained set like Cloud Translate do itās magic. Cloud Translate is a ādeep learningā process that has learned how to translate text to other languages a more natural (supernatural I should say) way.
