There is a strong correlation between the coins and also there is also a lag in the price change of one coin compared to others, this can be used to our advantage in prediction by using information from one coin to predict the target of another coin. I was thinking of using graph-based time-series predictions, has anyone already utilized this approach also maybe share your insights on whether the approach could be useful or not.
", "messages": [ { "id": "1651999", "postDate": "01/16/2022 08:52:54", "content": "There is a strong correlation between the coins and also there is also a lag in the price change of one coin compared to others, this can be used to our advantage in prediction by using information from one coin to predict the target of another coin. I was thinking of using graph-based time-series predictions, has anyone already utilized this approach also maybe share your insights on whether the approach could be useful or not.
", "rawMarkdown": "There is a strong correlation between the coins and also there is also a lag in the price change of one coin compared to others, this can be used to our advantage in prediction by using information from one coin to predict the target of another coin. I was thinking of using graph-based time-series predictions, has anyone already utilized this approach also maybe share your insights on whether the approach could be useful or not.", "votes": null }, { "id": "1656835", "postDate": "01/19/2022 17:03:55", "content": "I'm trying to use Graph Attention on this dataset, but sadly, it seems useless. My final corr is smaller than 0.
", "rawMarkdown": "I'm trying to use Graph Attention on this dataset, but sadly, it seems useless. My final corr is smaller than 0.", "votes": null }, { "id": "1659885", "postDate": "01/22/2022 07:16:20", "content": "The performance with a simple STGCN is also not too promising. I guess it has got to do with the ground truth generation as it destroys the autoregression part of the time series and to an extent also destroys the relationship between different coins. I still believe this would be a better approach with the GCN's rather than individual coin models, where mapping interactions between coins can be very difficult to model
\n@jxxxxxxxxx How did you create your graph?
", "rawMarkdown": "The performance with a simple STGCN is also not too promising. I guess it has got to do with the ground truth generation as it destroys the autoregression part of the time series and to an extent also destroys the relationship between different coins. I still believe this would be a better approach with the GCN's rather than individual coin models, where mapping interactions between coins can be very difficult to model\n\n@jxxxxxxxxx How did you create your graph?", "votes": null }, { "id": "1659955", "postDate": "01/22/2022 08:58:31", "content": "Just set edges for the nodes in same time period, and GAT can automally calculate the weight between these nodes due to the attention mechanism.
", "rawMarkdown": "Just set edges for the nodes in same time period, and GAT can automally calculate the weight between these nodes due to the attention mechanism.", "votes": null } ], "comments": [ { "id": 1656835, "author_name": "jxxxxxxxxx", "author_url": "", "post_date": "01/19/2022 17:03:55", "content": "I'm trying to use Graph Attention on this dataset, but sadly, it seems useless. My final corr is smaller than 0.
", "votes": null, "replies": [] }, { "id": 1659885, "author_name": "ssk619", "author_url": "", "post_date": "01/22/2022 07:16:20", "content": "The performance with a simple STGCN is also not too promising. I guess it has got to do with the ground truth generation as it destroys the autoregression part of the time series and to an extent also destroys the relationship between different coins. I still believe this would be a better approach with the GCN's rather than individual coin models, where mapping interactions between coins can be very difficult to model
\n@jxxxxxxxxx How did you create your graph?
", "votes": null, "replies": [ { "id": 1659955, "author_name": "jxxxxxxxxx", "author_url": "", "post_date": "01/22/2022 08:58:31", "content": "Just set edges for the nodes in same time period, and GAT can automally calculate the weight between these nodes due to the attention mechanism.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1651999": "There is a strong correlation between the coins and also there is also a lag in the price change of one coin compared to others, this can be used to our advantage in prediction by using information from one coin to predict the target of another coin. I was thinking of using graph-based time-series predictions, has anyone already utilized this approach also maybe share your insights on whether the approach could be useful or not.", "1656835": "I'm trying to use Graph Attention on this dataset, but sadly, it seems useless. My final corr is smaller than 0.", "1659885": "The performance with a simple STGCN is also not too promising. I guess it has got to do with the ground truth generation as it destroys the autoregression part of the time series and to an extent also destroys the relationship between different coins. I still believe this would be a better approach with the GCN's rather than individual coin models, where mapping interactions between coins can be very difficult to model\n\n@jxxxxxxxxx How did you create your graph?", "1659955": "Just set edges for the nodes in same time period, and GAT can automally calculate the weight between these nodes due to the attention mechanism." }, "source": "meta" }