# Getting started as Developer Check out the link to our Google Colab to get started with Crypticorn's HiveAI. {% embed url="" fullWidth="false" %} ## Overview ### Introduction HiveAI leverages the swarm intelligence of thousands of developers to create the best predictions for cryptocurrencies. You are provided with free, high quality data that you can use to train models and evaluate your predictions. Well performing models will be available for contributing to the Meta Model ("Hive AI") and rewarded in AIC. You can [sign up](https://numer.ai/signup) and visit your [dashboard](https://app.crypticorn.com/developer) for a full suite of tutorials. While reading the documentation, you should follow along the [notebook tutorial](https://app.crypticorn.dev/developer/docs/crypticorn.html) to work with code snippets and crypticorn's python client. All useful links and ressources are accessible through the `Help` tab in the dashboard ### Models Your objective is to build machine learning models to predict the `target` given the `features`. You can use any language or framework that you like. > Before you can download data you need to initialize a model. You can either do this in the dashboard or [directly with the python client](https://app.crypticorn.dev/developer/docs/crypticorn.html#Crypticorn.create_model). This process is mainly referred to as ***creating a model***. You create a model for a specific `coin` and `target` right from the beginning. Don't make the decision to complicated what to choose for these parameters. If you want to train your model on different coin or target, simply create a new model. To get an overview about available options to choose from you can use the [python client](https://app.crypticorn.dev/developer/docs/crypticorn.html#Crypticorn.data_info) ### Data Crypticorns's free datasets are made of multiple data sources, ranging from sentiment, news, OHLCV, Google Trends to several indicators. The dataset is ***obfuscated*** so that it can be given out for free and modeled without any financial domain knowledge. This also means that models you build on this data can only be used with Crypticorn's data. > Features are quantitative attributes known about a cryptocurrency at a time (e.g Open Price, SMA, search volume, etc.). Targets are measures of a cryptocurrency at a time that are to be predicted (e.g. the exact price). The data consists of multiple datasets for a range of coins. The datasets for each coin are split into `X` and `y` sets, `X` containing the `features`, `y` the `target`. The rows represent the value for each feature or target at a specific time. You will receive three files on download: `X_train`, `X_test` and `y_train` (under path e.g. *v1.0/coin\_1/* in your environment) You can download different sizes for the feature datasets, to match the needs for your development workflow. You might want to start with downloading the smallest feature set and finetune your model with bigger feature sets later on. You can download data for a specific model, specific version (optional) and specific feature size (optional) both via the `Data` tab in the dashboard or the [python client](https://app.crypticorn.dev/developer/docs/crypticorn.html#Crypticorn.download_data) ### Evaluation At this point you have started with using the downloaded data to train you model on a specific coin, predicting the target. Once you have your first prediction ready you can start evaluating you model by submitting your prediction to our backend and receive a evaluation. This step can only be done with the [python client](https://app.crypticorn.dev/developer/docs/crypticorn.html#Crypticorn.evaluate_model). The backend will return a response containing standard machine learning metrics depending on the target type (binary or continuous). The response also includes a naive and a random benchmark for each metric, which your model should outperform. The metrics used for the leaderboard ranking are the row-wise correlation (`CORR`) for continuous targets and the accuracy score (`ACS`) for binary targets. Both these values are own computations. ### Model Upload // WORK IN PROGRESS Well performing models will be available for submission and admitted to Hive AI ### Leaderboard The leaderboard categorizes the models by target type and ranks the models by the latest evaluation score of `CORR` or `ACS`. The leaderboard is grouped into season, each of them being bound to a data version (e.g. Season 1 lasts from the release of v1.0 to the release of v2.0), making sure that all the models within a season have the same conditions. You can view the [current leaderboard](https://app.crypticorn.dev/developer/leaderboard) and all [historic seasons](https://app.crypticorn.dev/developer/leaderboard/seasons). ### Rewards // WORK IN PROGRESS Submitted models/Models admitted to Hive AI will be rewarded in AIC token. ### Live Data // WORK IN PROGRESS Don't want to contribute to this ecosystem, but access our high quality data? Pay in AIC to get live data to integrate in your own trading bot. ### Bot Marketplace // WORK IN PROGRESS The 3rd Bot Marketplace is the integration of other developers, quants, and institutions’ trading bots into the Crypticorn ecosystem. This trading bot utilizes the Crypticorn data pipeline, predictions, and trade execution systems. This makes it easy for the developers to create strategies and then allow them to trade. Crypticorn will test and evaluate these trading strategies before token holders can select them and allocate funds to make them. This gives all Crypticorn token holders access to a wider range of strategies. The performance fee model also accounts for these bots, and the developer decides on the height of the performance fees from the user. Depending on the trading strategy's complexity and resource usage, the developer earns B% of the performance fees, while the remaining A% of the performance fees go to Crypticorn for the use of our technology and trade execution.