prediction vs labels
Opened this issue · 0 comments
Upon using python train.py lstnet solar_energy --context_points 168 --target_points 24 --run_name spatiotemporal_al_solar --batch_size 25:
test/loss 0.11
test/mae. 1.82
test/mape 19922692
test/mse 12.67
test/norm_mae 0.18
test/norm_mse. 0.11
test/smape. 1.417
question1: why the mape is so big? are these results correct?
I wanted to get the predictions and manually compare them with the labels using the following code:
forecaster.eval()
test_dataloader = data_module.test_dataloader()
for batch in test_dataloader:
# Extract input features from the batch
xc, yc, xt, yt = batch # Assuming these are the keys in your dataset
# Make predictions using the forecaster
yt_pred = forecaster.predict(xc, yc, xt,yt)
print(yt_pred)
print(yt)
break
This is the output:
`tensor([[[ 0.2024, 0.0527, 0.0133, ..., -0.0635, 0.0884, -0.0149],
[ 0.2010, 0.0525, 0.0131, ..., -0.0636, 0.0879, -0.0150],
[ 0.2041, 0.0523, 0.0131, ..., -0.0637, 0.0892, -0.0148],
...,
[ 0.7595, 0.0227, 0.0320, ..., -0.0812, 0.2956, 0.0085],
[ 0.8442, 0.0236, 0.0376, ..., -0.0806, 0.3280, 0.0181],
[ 0.8337, 0.0235, 0.0369, ..., -0.0809, 0.3239, 0.0163]],
[[ 0.2031, 0.0525, 0.0132, ..., -0.0635, 0.0887, -0.0148],
[ 0.2030, 0.0523, 0.0131, ..., -0.0636, 0.0887, -0.0148],
[ 0.2059, 0.0520, 0.0130, ..., -0.0638, 0.0898, -0.0146],
...,
[ 0.8137, 0.0230, 0.0353, ..., -0.0811, 0.3162, 0.0142],
[ 0.8793, 0.0247, 0.0404, ..., -0.0798, 0.3414, 0.0224],
[ 0.8796, 0.0252, 0.0408, ..., -0.0797, 0.3416, 0.0229]],
[[ 0.2051, 0.0523, 0.0131, ..., -0.0635, 0.0896, -0.0146],
[ 0.2048, 0.0520, 0.0129, ..., -0.0637, 0.0893, -0.0147],
[ 0.2073, 0.0514, 0.0128, ..., -0.0639, 0.0901, -0.0150],
...,
[ 0.8487, 0.0240, 0.0380, ..., -0.0805, 0.3296, 0.0183],
[ 0.9288, 0.0267, 0.0446, ..., -0.0783, 0.3606, 0.0295],
[ 0.9326, 0.0263, 0.0444, ..., -0.0791, 0.3620, 0.0290]],
...,
[[ 0.4881, 0.0286, 0.0199, ..., -0.0763, 0.1952, -0.0072],
[ 0.5514, 0.0270, 0.0233, ..., -0.0777, 0.2192, -0.0024],
[ 0.6315, 0.0262, 0.0280, ..., -0.0786, 0.2497, 0.0051],
...,
[ 0.8561, 0.0438, 0.0564, ..., -0.0773, 0.3446, 0.0408],
[ 0.8403, 0.0452, 0.0564, ..., -0.0763, 0.3386, 0.0404],
[ 0.8614, 0.0461, 0.0588, ..., -0.0754, 0.3468, 0.0431]],
[[ 0.5504, 0.0270, 0.0233, ..., -0.0778, 0.2188, -0.0025],
[ 0.6529, 0.0259, 0.0291, ..., -0.0788, 0.2578, 0.0069],
[ 0.7530, 0.0256, 0.0349, ..., -0.0795, 0.2957, 0.0162],
...,
[ 0.8595, 0.0458, 0.0580, ..., -0.0759, 0.3459, 0.0428],
[ 0.8386, 0.0458, 0.0571, ..., -0.0757, 0.3378, 0.0406],
[ 0.8749, 0.0465, 0.0601, ..., -0.0752, 0.3520, 0.0446]],
[[ 0.6496, 0.0259, 0.0290, ..., -0.0789, 0.2564, 0.0067],
[ 0.7954, 0.0254, 0.0371, ..., -0.0797, 0.3117, 0.0198],
[ 0.8922, 0.0258, 0.0428, ..., -0.0797, 0.3481, 0.0284],
...,
[ 0.8682, 0.0466, 0.0594, ..., -0.0751, 0.3492, 0.0442],
[ 0.8442, 0.0460, 0.0578, ..., -0.0757, 0.3398, 0.0411],
[ 0.8905, 0.0470, 0.0614, ..., -0.0748, 0.3581, 0.0463]]])
tensor([[[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
...,
[ 0.9684, 0.1093, 0.3574, ..., 0.3511, 1.0403, 0.2583],
[ 1.0116, 0.2213, 0.5154, ..., 0.5210, 1.1411, 0.4755],
[ 1.0467, 0.4340, 0.6848, ..., 0.6800, 1.2230, 0.7388]],
[[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
...,
[ 1.0116, 0.2213, 0.5154, ..., 0.5210, 1.1411, 0.4755],
[ 1.0467, 0.4340, 0.6848, ..., 0.6800, 1.2230, 0.7388],
[ 1.0521, 0.6411, 0.8429, ..., 0.7512, 1.2734, 0.9018]],
[[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
[-0.8025, -0.7136, -0.6984, ..., -0.7011, -0.7864, -0.6945],
...,
[ 1.0467, 0.4340, 0.6848, ..., 0.6800, 1.2230, 0.7388],
[ 1.0521, 0.6411, 0.8429, ..., 0.7512, 1.2734, 0.9018],
[ 1.0251, 0.7811, 0.9897, ..., 0.8718, 1.3616, 1.1442]],
...,
[[ 1.0116, 0.2213, 0.5154, ..., 0.5210, 1.1411, 0.4755],
[ 1.0467, 0.4340, 0.6848, ..., 0.6800, 1.2230, 0.7388],
[ 1.0521, 0.6411, 0.8429, ..., 0.7512, 1.2734, 0.9018],
...,
[ 1.2059, 1.8111, 2.0511, ..., 1.9898, 1.2419, 2.0760],
[ 0.7632, 1.6935, 2.0793, ..., 2.0446, 1.2923, 2.1972],
[ 1.0710, 1.5648, 2.1019, ..., 2.0720, 1.3427, 2.1429]],
[[ 1.0467, 0.4340, 0.6848, ..., 0.6800, 1.2230, 0.7388],
[ 1.0521, 0.6411, 0.8429, ..., 0.7512, 1.2734, 0.9018],
[ 1.0251, 0.7811, 0.9897, ..., 0.8718, 1.3616, 1.1442],
...,
[ 0.7632, 1.6935, 2.0793, ..., 2.0446, 1.2923, 2.1972],
[ 1.0710, 1.5648, 2.1019, ..., 2.0720, 1.3427, 2.1429],
[ 0.9198, 1.2289, 2.1414, ..., 2.0884, 1.3112, 2.1763]],
[[ 1.0521, 0.6411, 0.8429, ..., 0.7512, 1.2734, 0.9018],
[ 1.0251, 0.7811, 0.9897, ..., 0.8718, 1.3616, 1.1442],
[ 0.5824, 0.6019, 1.1195, ..., 1.0417, 1.4183, 1.1358],
...,
[ 1.0710, 1.5648, 2.1019, ..., 2.0720, 1.3427, 2.1429],
[ 0.9198, 1.2289, 2.1414, ..., 2.0884, 1.3112, 2.1763],
[ 1.1438, 1.1449, 2.1470, ..., 2.0830, 1.2797, 2.2348]]])`
question2: In AL-solar, there are no negatives but the labels have some negative values? are the some scaling going on in here?
question3: even then my predictions are not at all near to the labels (I am using the provided prediction method), any idea why?