I have a situation where multiple inputs (rows) makes up the single output result. My problem is similar to this one: How to train a machine/deep learning model with data that has multiple input rows per day but only one row of label/ground-truth (output) data per day but does not refer to time series, only spatial data.
Let say I want to predict a value of imision in some central points. Each of them depends on a set of margin points (emitters) and their properties like distance, emission rate, topographic obstacles. Each connection between central and margin points constitutes one row, so for the single central point, I will have many rows of data.
It gives a table form of
<style>
th, td {
border:1px solid black;
text-align:center;
}
</style>
<table>
<tr>
<th>inp1</th>
<th>inp2</th>
<th>...</th>
<th>inpn</th>
<th>Output</th>
</tr>
<tr>
<td>x11</td>
<td>x12</td>
<td>...</td>
<td>x1n</td>
<td rowspan=4>y1</td>
</tr>
<tr>
<td>x21</td>
<td>x22</td>
<td>...</td>
<td>x2n</td>
</tr>
<tr>
<td>...</td>
<td>...</td>
<td>...</td>
<td>...</td>
</tr>
<tr>
<td>xm1</td>
<td>xm2</td>
<td>...</td>
<td>xmn</td>
</tr>
<tr>
<td>x201</td>
<td>x202</td>
<td>...</td>
<td>x20n</td>
<td rowspan=4>y2</td>
</tr>
<tr>
<td>...</td>
<td>...</td>
<td>...</td>
<td>...</td>
</tr>
</table>For example:
|no| distance| emission rate| substance| ....|
1| 100|5.44| aaa | ....|
2| 100|5.44| aaa | ....|
Both rows 1 and 2 make the single output of some y1. If only one row of the above was considered the output would be different than the result with two rows (!=y1). Then other rows (3, 4, 5, 6) make another single output of y2. And so on.
Are there any deep/ machine learning solutions/ topologies for this? I know that the topic has already been discussed, but the only solutions I found was to concatenate many rows into one or to aggregate values of particular parameters (like sum, median, etc) which may be insufficient or impossible in my case.
I'm wondering whether there is something new in this topic, especially since my case is not the only one.