Uploaded December 2023 | Updated September 2026, 2 hours ago

---

I trained a neural network on weather data of a lake in Switzerland. In this video, I try to visualize its predictions to get a better intuition of what the model does and how the different features contribute.

In case you are interested: this is the flax snippet of the model architecture:

```
def __call__(self, x):
x = nn.Conv(features=self.channels, kernel_size=(1, self.conv_len))(x)
x = nn.sigmoid(x)
x = nn.max_pool(x, window_shape=(1, self.down_scale), strides=(1, self.down_scale))
x = nn.Conv(features=self.channels, kernel_size=(1, self.conv_len))(x)
x = nn.sigmoid(x)
x = nn.max_pool(x, window_shape=(1, self.down_scale), strides=(1, self.down_scale))
x = x.reshape((x.shape[0], -1)) # flatten
x = nn.Dense(features=self.dense_size)(x)
x = nn.sigmoid(x)
x = nn.Dense(features=self.predictions)(x)
x = nn.sigmoid(x)
return x
```

The data is sampled at 10 minute intervals.

Parameters:
--history=257
--conv_len=16
--down_scale=2
--channels=20
--dense_size=250
--learning_rate=0.001
--batch_size=64
--predictions=120