Intro to the Platform #back to top

The Vemity platform is intended to guide even the most inexperienced developers through using AI. The only required knowledge is that of accessing a RESTful API. The three primary steps are outlined below.

Create

From the user dashboard, click the New Model button on the top right corner of the page.

• Step 1 - Choose what kind of AI you want to build.
• Step 2 - Upload your dataset or find one of ours to use.
• Step 3 - Tell us a little bit about your dataset and what you want to do.

Train

At this point, you have built the model. Until the AI has trained, it isn't any good. You have just a few more steps.

• Step 4 - Choose which kind of training method to use.
• Step 5 - Set the number of training epochs that best fit your budget.
• Step 6 - Train! Once it's done, you will be able to see the results and retrain if needed.

Connect

From here, your AI is built, but now what? This is where you leverage the awesome power of AI.

• Step 7 - Click the view button next to the model you want to inference.
• Step 8 - Use the information on this page to set up your API request.
• Step 9 - Send new data to your AI through our API and get predictions / classifications quickly.

How to find Datasets? #back to top

Datasets are what make or break an AI. In general there are a few ways to go about getting a dataset for what you need.

Choose from a collection of datasets we have ready to use. When setting up your model, choose the "Browse our datasets" option.

A lot of universities and public organizations have great datasets that are available through their websites. Sometimes enough searching will lead to finding the perfect dataset. Or at least one that you can slightly manipulate. Also take a look at datasets on Kaggle.

There are a number of ways to do this! You could set up web-scraping scripts to pull data from the internet; you could interview a large number of people; you could find a few similar datasets and merge them into one that fits your needs; or you can harvest data from your existing users / customers.

Some datasets are just not available for free and you might not have the ability to build it yourself. Organizations and businesses will sometimes sell you their data, but this is typically a B2B transaction and we can't really help you with that ¯\_(ツ)_/¯

Data Classifier Demo #back to top

Image Classifier Demo #back to top

We're working on getting this done! It'll be here soon.

There are three different methods for training your AI.

Manual Training

Manual training works exactly how it sounds. You choose exactly how many epochs (loops through the entire dataset) to train for and what learning rate (how much impact each step of training has) to use. We do exactly what you said and return the trained model. This option gives you a lot of control over your final training fee, but can lead to over-training or under-training.

Optimal Training

Optimal training will train the AI until it's done. It will continue training until the accuracy is not continuing to increase. At this point, it cuts it off and returns the trained AI. That returned AI is likely the best it could be for that training session. For this reason, optimal training is better than manual training if you're not familiar with AI. To prevent your training session from running over-budget, we have provided an option to cap your epochs. With this setting, you can use optimal training, but stop it at a set number of epochs. This way, if an AI only needs 8 epochs to be accurate, you only get charged for 8. However, you can also protect yourself from over-spending on training if it actually needed 120, but you only wanted to spend the credits needed for 10.

Repetitive Optimal Training

Optimal training provides the best accuracy on a static neural network model in a single training session. Repetitive Optimal Training will run 5 training sessions in optimal training mode. It will train the first AI, see how accurate it is, then slightly adjust the model to get better performance on the next training session. It does this 5 times, until it finds a very accurate AI for your dataset. NOTE: Repetitive Optimal Training will charge 5x the credits as Optimal, however you can still cap the epochs as desired.

Custom Models

Data Classifier #back to top

Step 1 - Get your API key and Model ID from the AI you want to inference. Do this by accessing your Vemity Dashboard and viewing your AI.
Step 2 - Configure your program to send a HTTP POST request with a json body.
Step 3a - Turn your input data into a two dimensional array. If you only have one input, it will look like [["first","data"]]. If you have more than one input, format it as [["first","data"], ["second","data"],...]. The AI will return predictions for each sub-array in the array.
Step 3b - Turn your input data into an array of dictionaries. If you only have one input, it will look like [{"label":"value", "label2":"value2"}]. If you have more than one input, format it as [{"label":"value", "label2":"value2"}, {"foo":"bar", "fee":"foh"},...]. The keys to the dictionary must match the keys on the model's view page (they do not need to be in order). The AI will return predictions for each dictionary in the array.
Step 4 - Format your json body as such:

 Multi-Dimensional Array

                                    {
                                      id: "YOUR_MODELS_ID",
                                      input: [["input","data"],["as","a","2D","array"]],
                                      email: "YOUR_EMAIL",
                                      key: "YOUR_API_KEY",
                                    }
                                                                                                                        .
                                  

 Dictionary Array

                                    {
                                      id: "YOUR_MODELS_ID",
                                      input: [{
                                        "foo":"bar",
                                        "fee":"foh"
                                      }],
                                      email: "YOUR_EMAIL",
                                      key: "YOUR_API_KEY",
                                    }
                                                                                                                        .
                                  

Image Classifier #back to top

Step 1 - Get your API key and Model ID from the AI you want to inference. Do this by accessing your Vemity Dashboard and viewing your AI.
Step 2 - Configure your program to send a HTTP POST request with a json body.
Step 3 - Convert your image input into base64 format. Make sure to keep the data:image/jpeg;base64, at the beginning of the string.
NOTE: Images are inferenced one at a time. You cannot send multiple images in one request.
DOUBLE NOTE: In order to speed up inference time for yourself and other users, it is recommended that you reduce image sizes to a max of 250px by 250px. The AI will do this anyway, so you won't miss out on potential accuracy by do this.
Step 4 - Format your json body as such:

                                    {
                                      id: "YOUR_MODELS_ID",
                                      input: "BASE64 ENCODED STRING",
                                      email: "YOUR_EMAIL",
                                      key: "YOUR_API_KEY",
                                    }
                                                                                                                        .
                                  

Static Models

These are generic models that are available to Vemity users for common use-cases. These models operate similarly to custom models.

Twitter Sentiment Analysis #back to top

Step 1 - Retrieve your API key from your account page or any model view page.
Step 2 - Configure your program to send a HTTP POST request with a json body.
Step 3 - Turn your input tweets into an array. If you only have one input, it will look like ["tweet"]. If you have more than one input, format it as ["tweet", "tweet",...]. The AI will return predictions for each tweet in the array.
Step 4 - Format your json body as such:

                                    {
                                      input: ["TWEET_ARRAY"],
                                      email: "YOUR_EMAIL",
                                      key: "YOUR_API_KEY",
                                    }
                                                                                                                        .
                                  

Credit System

In order to support a "pay-as-you-go" model, Vemity uses a credit system to manage AI usage.
1 credit is worth $0.005. Users purchase a number of credits to spend on training and inferencing.

Free Credits

At this point in time, all new Vemity users will receive 2,500 free credits upon creating an account and confirming their email.

Data Classification

* One training step is defined as one row in the dataset. If your dataset has 500 rows and you train for 5 epochs, you will be charged for 2,500 steps. This is 25 credits or $0.125

Image Classification

* One training step is defined as one picture. If your dataset has 500 pictures and you train for 5 epochs, you will be charged for 2,500 steps. This is 125 credits or $0.625

Credit Refil System

As you use the system, you will be charged credits automatically. As you begin to run low on credits, you will be asked to refill your credit balance. Once you refill, you will have the option to auto-refill upon reaching a pre-determined balance. This ensures your AI access will never be denied.

EXAMPLE: Once your balance reaches 200 credits or less, your account can automatically refill to a balance of 2,000.

NOTE: You do not need to enter a payment method until you refill from your original, free balance.

Submit any bug reports or support requests on our support contact page.