CS194 Proj 4¶
In [34]:
Out[34]:
('T-shirt/top',
'Trouser',
'Pullover',
'Dress',
'Coat',
'Sandal',
'Shirt',
'Sneaker',
'Bag',
'Ankle boot')
In [36]:
tensor([2, 0, 6, 2])
In [37]:
Pullover T-shirt/top Shirt Pullover
In [38]:
Credit given: This tutorial helped me learn how to use pytorch: https://adventuresinmachinelearning.com/convolutional-neural-networks-tutorial-in-pytorch/
DATALOADER
NET: I found that using 32 for the output channels of conv1 and then 32 input and output for conv2 layer worked well. Kernel size 3. The things that changed the accuracy the most were learning rate and the batch size. Batch size 20 seemed to be most effective anything higher or lower resulted in worse results. I initally used a 0.01 learning rate but found 0.001 to be best and anything higher or lower resulted in worse results. MaxPool had the typical (2,2) params.
In [53]:
Epoch [1/10], Step [100/3000], Loss: 0.4628, Accuracy: 85.00% Epoch [1/10], Step [200/3000], Loss: 0.6652, Accuracy: 80.00% Epoch [1/10], Step [300/3000], Loss: 0.5107, Accuracy: 80.00% Epoch [1/10], Step [400/3000], Loss: 0.1357, Accuracy: 95.00% Epoch [1/10], Step [500/3000], Loss: 0.4701, Accuracy: 85.00% Epoch [1/10], Step [600/3000], Loss: 0.5581, Accuracy: 85.00% Epoch [1/10], Step [700/3000], Loss: 0.2375, Accuracy: 90.00% Epoch [1/10], Step [800/3000], Loss: 0.1961, Accuracy: 95.00% Epoch [1/10], Step [900/3000], Loss: 0.2854, Accuracy: 90.00% Epoch [1/10], Step [1000/3000], Loss: 0.4166, Accuracy: 80.00% Epoch [1/10], Step [1100/3000], Loss: 0.5280, Accuracy: 85.00% Epoch [1/10], Step [1200/3000], Loss: 0.4247, Accuracy: 80.00% Epoch [1/10], Step [1300/3000], Loss: 0.2999, Accuracy: 85.00% Epoch [1/10], Step [1400/3000], Loss: 0.3213, Accuracy: 90.00% Epoch [1/10], Step [1500/3000], Loss: 0.5147, Accuracy: 80.00% Epoch [1/10], Step [1600/3000], Loss: 0.3819, Accuracy: 85.00% Epoch [1/10], Step [1700/3000], Loss: 0.4780, Accuracy: 85.00% Epoch [1/10], Step [1800/3000], Loss: 0.2692, Accuracy: 90.00% Epoch [1/10], Step [1900/3000], Loss: 0.3079, Accuracy: 85.00% Epoch [1/10], Step [2000/3000], Loss: 0.3034, Accuracy: 95.00% Epoch [1/10], Step [2100/3000], Loss: 0.5501, Accuracy: 70.00% Epoch [1/10], Step [2200/3000], Loss: 0.2512, Accuracy: 85.00% Epoch [1/10], Step [2300/3000], Loss: 0.4515, Accuracy: 85.00% Epoch [1/10], Step [2400/3000], Loss: 0.1739, Accuracy: 100.00% Epoch [1/10], Step [2500/3000], Loss: 0.8296, Accuracy: 70.00% Epoch [1/10], Step [2600/3000], Loss: 0.2066, Accuracy: 85.00% Epoch [1/10], Step [2700/3000], Loss: 0.2910, Accuracy: 90.00% Epoch [1/10], Step [2800/3000], Loss: 0.2869, Accuracy: 80.00% Epoch [1/10], Step [2900/3000], Loss: 0.2939, Accuracy: 85.00% Epoch [1/10], Step [3000/3000], Loss: 0.5685, Accuracy: 75.00% Epoch [2/10], Step [100/3000], Loss: 0.3273, Accuracy: 90.00% Epoch [2/10], Step [200/3000], Loss: 0.1861, Accuracy: 95.00% Epoch [2/10], Step [300/3000], Loss: 0.0400, Accuracy: 100.00% Epoch [2/10], Step [400/3000], Loss: 0.3638, Accuracy: 80.00% Epoch [2/10], Step [500/3000], Loss: 0.5016, Accuracy: 90.00% Epoch [2/10], Step [600/3000], Loss: 0.2348, Accuracy: 90.00% Epoch [2/10], Step [700/3000], Loss: 0.3424, Accuracy: 80.00% Epoch [2/10], Step [800/3000], Loss: 0.3965, Accuracy: 80.00% Epoch [2/10], Step [900/3000], Loss: 0.7621, Accuracy: 75.00% Epoch [2/10], Step [1000/3000], Loss: 0.3851, Accuracy: 80.00% Epoch [2/10], Step [1100/3000], Loss: 0.2229, Accuracy: 95.00% Epoch [2/10], Step [1200/3000], Loss: 0.2413, Accuracy: 85.00% Epoch [2/10], Step [1300/3000], Loss: 0.2122, Accuracy: 90.00% Epoch [2/10], Step [1400/3000], Loss: 0.3876, Accuracy: 85.00% Epoch [2/10], Step [1500/3000], Loss: 0.2901, Accuracy: 95.00% Epoch [2/10], Step [1600/3000], Loss: 0.3737, Accuracy: 85.00% Epoch [2/10], Step [1700/3000], Loss: 0.3172, Accuracy: 90.00% Epoch [2/10], Step [1800/3000], Loss: 0.0848, Accuracy: 100.00% Epoch [2/10], Step [1900/3000], Loss: 0.1744, Accuracy: 90.00% Epoch [2/10], Step [2000/3000], Loss: 0.1906, Accuracy: 95.00% Epoch [2/10], Step [2100/3000], Loss: 0.4993, Accuracy: 85.00% Epoch [2/10], Step [2200/3000], Loss: 0.2163, Accuracy: 85.00% Epoch [2/10], Step [2300/3000], Loss: 0.2635, Accuracy: 90.00% Epoch [2/10], Step [2400/3000], Loss: 0.2485, Accuracy: 95.00% Epoch [2/10], Step [2500/3000], Loss: 0.3408, Accuracy: 85.00% Epoch [2/10], Step [2600/3000], Loss: 0.4319, Accuracy: 80.00% Epoch [2/10], Step [2700/3000], Loss: 0.6116, Accuracy: 80.00% Epoch [2/10], Step [2800/3000], Loss: 0.1679, Accuracy: 95.00% Epoch [2/10], Step [2900/3000], Loss: 0.0911, Accuracy: 95.00% Epoch [2/10], Step [3000/3000], Loss: 0.1817, Accuracy: 95.00% Epoch [3/10], Step [100/3000], Loss: 0.3733, Accuracy: 85.00% Epoch [3/10], Step [200/3000], Loss: 0.3216, Accuracy: 90.00% Epoch [3/10], Step [300/3000], Loss: 0.3410, Accuracy: 90.00% Epoch [3/10], Step [400/3000], Loss: 0.2282, Accuracy: 90.00% Epoch [3/10], Step [500/3000], Loss: 0.3630, Accuracy: 85.00% Epoch [3/10], Step [600/3000], Loss: 0.1932, Accuracy: 95.00% Epoch [3/10], Step [700/3000], Loss: 0.1167, Accuracy: 95.00% Epoch [3/10], Step [800/3000], Loss: 0.4829, Accuracy: 85.00% Epoch [3/10], Step [900/3000], Loss: 0.1845, Accuracy: 90.00% Epoch [3/10], Step [1000/3000], Loss: 0.1607, Accuracy: 95.00% Epoch [3/10], Step [1100/3000], Loss: 0.4644, Accuracy: 85.00% Epoch [3/10], Step [1200/3000], Loss: 0.3389, Accuracy: 95.00% Epoch [3/10], Step [1300/3000], Loss: 0.3985, Accuracy: 80.00% Epoch [3/10], Step [1400/3000], Loss: 0.1940, Accuracy: 95.00% Epoch [3/10], Step [1500/3000], Loss: 0.2237, Accuracy: 95.00% Epoch [3/10], Step [1600/3000], Loss: 0.1893, Accuracy: 90.00% Epoch [3/10], Step [1700/3000], Loss: 0.4077, Accuracy: 80.00% Epoch [3/10], Step [1800/3000], Loss: 0.2109, Accuracy: 90.00% Epoch [3/10], Step [1900/3000], Loss: 0.7823, Accuracy: 75.00% Epoch [3/10], Step [2000/3000], Loss: 0.1997, Accuracy: 90.00% Epoch [3/10], Step [2100/3000], Loss: 0.0800, Accuracy: 100.00% Epoch [3/10], Step [2200/3000], Loss: 0.3059, Accuracy: 90.00% Epoch [3/10], Step [2300/3000], Loss: 0.0912, Accuracy: 95.00% Epoch [3/10], Step [2400/3000], Loss: 0.4567, Accuracy: 90.00% Epoch [3/10], Step [2500/3000], Loss: 0.1343, Accuracy: 95.00% Epoch [3/10], Step [2600/3000], Loss: 0.0339, Accuracy: 100.00% Epoch [3/10], Step [2700/3000], Loss: 0.1620, Accuracy: 90.00% Epoch [3/10], Step [2800/3000], Loss: 0.3142, Accuracy: 85.00% Epoch [3/10], Step [2900/3000], Loss: 0.3319, Accuracy: 90.00% Epoch [3/10], Step [3000/3000], Loss: 0.1551, Accuracy: 95.00% Epoch [4/10], Step [100/3000], Loss: 0.1485, Accuracy: 95.00% Epoch [4/10], Step [200/3000], Loss: 0.2251, Accuracy: 85.00% Epoch [4/10], Step [300/3000], Loss: 0.1974, Accuracy: 95.00% Epoch [4/10], Step [400/3000], Loss: 0.5507, Accuracy: 80.00% Epoch [4/10], Step [500/3000], Loss: 0.4005, Accuracy: 80.00% Epoch [4/10], Step [600/3000], Loss: 0.0233, Accuracy: 100.00% Epoch [4/10], Step [700/3000], Loss: 0.1739, Accuracy: 95.00% Epoch [4/10], Step [800/3000], Loss: 0.2546, Accuracy: 90.00% Epoch [4/10], Step [900/3000], Loss: 0.1362, Accuracy: 95.00% Epoch [4/10], Step [1000/3000], Loss: 0.3317, Accuracy: 85.00% Epoch [4/10], Step [1100/3000], Loss: 0.1828, Accuracy: 95.00% Epoch [4/10], Step [1200/3000], Loss: 0.3769, Accuracy: 90.00% Epoch [4/10], Step [1300/3000], Loss: 0.0679, Accuracy: 100.00% Epoch [4/10], Step [1400/3000], Loss: 0.6137, Accuracy: 85.00% Epoch [4/10], Step [1500/3000], Loss: 0.0698, Accuracy: 100.00% Epoch [4/10], Step [1600/3000], Loss: 0.3762, Accuracy: 85.00% Epoch [4/10], Step [1700/3000], Loss: 0.4510, Accuracy: 70.00% Epoch [4/10], Step [1800/3000], Loss: 0.1885, Accuracy: 90.00% Epoch [4/10], Step [1900/3000], Loss: 0.5116, Accuracy: 80.00% Epoch [4/10], Step [2000/3000], Loss: 0.4353, Accuracy: 85.00% Epoch [4/10], Step [2100/3000], Loss: 0.3314, Accuracy: 90.00% Epoch [4/10], Step [2200/3000], Loss: 0.1867, Accuracy: 90.00% Epoch [4/10], Step [2300/3000], Loss: 0.1656, Accuracy: 95.00% Epoch [4/10], Step [2400/3000], Loss: 0.2135, Accuracy: 90.00% Epoch [4/10], Step [2500/3000], Loss: 0.0646, Accuracy: 100.00% Epoch [4/10], Step [2600/3000], Loss: 0.2561, Accuracy: 95.00% Epoch [4/10], Step [2700/3000], Loss: 0.5477, Accuracy: 75.00% Epoch [4/10], Step [2800/3000], Loss: 0.1750, Accuracy: 95.00% Epoch [4/10], Step [2900/3000], Loss: 0.1188, Accuracy: 95.00% Epoch [4/10], Step [3000/3000], Loss: 0.2354, Accuracy: 90.00% Epoch [5/10], Step [100/3000], Loss: 0.1843, Accuracy: 90.00% Epoch [5/10], Step [200/3000], Loss: 0.2618, Accuracy: 90.00% Epoch [5/10], Step [300/3000], Loss: 0.4223, Accuracy: 85.00% Epoch [5/10], Step [400/3000], Loss: 0.2884, Accuracy: 85.00% Epoch [5/10], Step [500/3000], Loss: 0.1124, Accuracy: 95.00% Epoch [5/10], Step [600/3000], Loss: 0.0763, Accuracy: 100.00% Epoch [5/10], Step [700/3000], Loss: 0.1257, Accuracy: 95.00% Epoch [5/10], Step [800/3000], Loss: 0.2191, Accuracy: 90.00% Epoch [5/10], Step [900/3000], Loss: 0.3318, Accuracy: 85.00% Epoch [5/10], Step [1000/3000], Loss: 0.2622, Accuracy: 90.00% Epoch [5/10], Step [1100/3000], Loss: 0.4539, Accuracy: 80.00% Epoch [5/10], Step [1200/3000], Loss: 0.1606, Accuracy: 95.00% Epoch [5/10], Step [1300/3000], Loss: 0.2890, Accuracy: 80.00% Epoch [5/10], Step [1400/3000], Loss: 0.0620, Accuracy: 100.00% Epoch [5/10], Step [1500/3000], Loss: 0.2721, Accuracy: 95.00% Epoch [5/10], Step [1600/3000], Loss: 0.1640, Accuracy: 90.00% Epoch [5/10], Step [1700/3000], Loss: 0.4831, Accuracy: 85.00% Epoch [5/10], Step [1800/3000], Loss: 0.2415, Accuracy: 95.00% Epoch [5/10], Step [1900/3000], Loss: 0.1434, Accuracy: 95.00% Epoch [5/10], Step [2000/3000], Loss: 0.1723, Accuracy: 90.00% Epoch [5/10], Step [2100/3000], Loss: 0.1722, Accuracy: 95.00% Epoch [5/10], Step [2200/3000], Loss: 0.0515, Accuracy: 100.00% Epoch [5/10], Step [2300/3000], Loss: 0.0601, Accuracy: 100.00% Epoch [5/10], Step [2400/3000], Loss: 0.4366, Accuracy: 85.00% Epoch [5/10], Step [2500/3000], Loss: 0.2367, Accuracy: 95.00% Epoch [5/10], Step [2600/3000], Loss: 0.5193, Accuracy: 95.00% Epoch [5/10], Step [2700/3000], Loss: 0.3482, Accuracy: 90.00% Epoch [5/10], Step [2800/3000], Loss: 0.2202, Accuracy: 95.00% Epoch [5/10], Step [2900/3000], Loss: 0.1114, Accuracy: 95.00% Epoch [5/10], Step [3000/3000], Loss: 0.3023, Accuracy: 90.00% Epoch [6/10], Step [100/3000], Loss: 0.2370, Accuracy: 90.00% Epoch [6/10], Step [200/3000], Loss: 0.1140, Accuracy: 95.00% Epoch [6/10], Step [300/3000], Loss: 0.0939, Accuracy: 95.00% Epoch [6/10], Step [400/3000], Loss: 0.2715, Accuracy: 90.00% Epoch [6/10], Step [500/3000], Loss: 0.2998, Accuracy: 85.00% Epoch [6/10], Step [600/3000], Loss: 0.0082, Accuracy: 100.00% Epoch [6/10], Step [700/3000], Loss: 0.1872, Accuracy: 90.00% Epoch [6/10], Step [800/3000], Loss: 0.0646, Accuracy: 100.00% Epoch [6/10], Step [900/3000], Loss: 0.1053, Accuracy: 95.00% Epoch [6/10], Step [1000/3000], Loss: 0.2477, Accuracy: 85.00% Epoch [6/10], Step [1100/3000], Loss: 0.3201, Accuracy: 85.00% Epoch [6/10], Step [1200/3000], Loss: 0.1981, Accuracy: 95.00% Epoch [6/10], Step [1300/3000], Loss: 0.0971, Accuracy: 100.00% Epoch [6/10], Step [1400/3000], Loss: 0.1975, Accuracy: 95.00% Epoch [6/10], Step [1500/3000], Loss: 0.2908, Accuracy: 90.00% Epoch [6/10], Step [1600/3000], Loss: 0.2407, Accuracy: 85.00% Epoch [6/10], Step [1700/3000], Loss: 0.4058, Accuracy: 85.00% Epoch [6/10], Step [1800/3000], Loss: 0.0579, Accuracy: 100.00% Epoch [6/10], Step [1900/3000], Loss: 0.1069, Accuracy: 95.00% Epoch [6/10], Step [2000/3000], Loss: 0.1596, Accuracy: 85.00% Epoch [6/10], Step [2100/3000], Loss: 0.2812, Accuracy: 80.00% Epoch [6/10], Step [2200/3000], Loss: 0.3675, Accuracy: 90.00% Epoch [6/10], Step [2300/3000], Loss: 0.1659, Accuracy: 95.00% Epoch [6/10], Step [2400/3000], Loss: 0.0171, Accuracy: 100.00% Epoch [6/10], Step [2500/3000], Loss: 0.3344, Accuracy: 95.00% Epoch [6/10], Step [2600/3000], Loss: 0.0610, Accuracy: 100.00% Epoch [6/10], Step [2700/3000], Loss: 0.1432, Accuracy: 95.00% Epoch [6/10], Step [2800/3000], Loss: 0.0608, Accuracy: 100.00% Epoch [6/10], Step [2900/3000], Loss: 0.0361, Accuracy: 100.00% Epoch [6/10], Step [3000/3000], Loss: 0.0456, Accuracy: 100.00% Epoch [7/10], Step [100/3000], Loss: 0.1034, Accuracy: 95.00% Epoch [7/10], Step [200/3000], Loss: 0.1703, Accuracy: 95.00% Epoch [7/10], Step [300/3000], Loss: 0.5123, Accuracy: 80.00% Epoch [7/10], Step [400/3000], Loss: 0.2785, Accuracy: 90.00% Epoch [7/10], Step [500/3000], Loss: 0.2047, Accuracy: 85.00% Epoch [7/10], Step [600/3000], Loss: 0.1411, Accuracy: 95.00% Epoch [7/10], Step [700/3000], Loss: 0.0936, Accuracy: 95.00% Epoch [7/10], Step [800/3000], Loss: 0.0746, Accuracy: 100.00% Epoch [7/10], Step [900/3000], Loss: 0.1848, Accuracy: 95.00% Epoch [7/10], Step [1000/3000], Loss: 0.0341, Accuracy: 100.00% Epoch [7/10], Step [1100/3000], Loss: 0.3269, Accuracy: 90.00% Epoch [7/10], Step [1200/3000], Loss: 0.4124, Accuracy: 90.00% Epoch [7/10], Step [1300/3000], Loss: 0.1099, Accuracy: 95.00% Epoch [7/10], Step [1400/3000], Loss: 0.1613, Accuracy: 95.00% Epoch [7/10], Step [1500/3000], Loss: 0.1032, Accuracy: 95.00% Epoch [7/10], Step [1600/3000], Loss: 0.3650, Accuracy: 90.00% Epoch [7/10], Step [1700/3000], Loss: 0.3603, Accuracy: 85.00% Epoch [7/10], Step [1800/3000], Loss: 0.2882, Accuracy: 90.00% Epoch [7/10], Step [1900/3000], Loss: 0.2114, Accuracy: 95.00% Epoch [7/10], Step [2000/3000], Loss: 0.0665, Accuracy: 100.00% Epoch [7/10], Step [2100/3000], Loss: 0.1988, Accuracy: 90.00% Epoch [7/10], Step [2200/3000], Loss: 0.0713, Accuracy: 100.00% Epoch [7/10], Step [2300/3000], Loss: 0.1868, Accuracy: 90.00% Epoch [7/10], Step [2400/3000], Loss: 0.2256, Accuracy: 90.00% Epoch [7/10], Step [2500/3000], Loss: 0.4054, Accuracy: 85.00% Epoch [7/10], Step [2600/3000], Loss: 0.2041, Accuracy: 95.00% Epoch [7/10], Step [2700/3000], Loss: 0.3682, Accuracy: 90.00% Epoch [7/10], Step [2800/3000], Loss: 0.1549, Accuracy: 95.00% Epoch [7/10], Step [2900/3000], Loss: 0.0913, Accuracy: 95.00% Epoch [7/10], Step [3000/3000], Loss: 0.1494, Accuracy: 95.00% Epoch [8/10], Step [100/3000], Loss: 0.2605, Accuracy: 90.00% Epoch [8/10], Step [200/3000], Loss: 0.2360, Accuracy: 85.00% Epoch [8/10], Step [300/3000], Loss: 0.1016, Accuracy: 95.00% Epoch [8/10], Step [400/3000], Loss: 0.1166, Accuracy: 100.00% Epoch [8/10], Step [500/3000], Loss: 0.0568, Accuracy: 100.00% Epoch [8/10], Step [600/3000], Loss: 0.5164, Accuracy: 95.00% Epoch [8/10], Step [700/3000], Loss: 0.3583, Accuracy: 90.00% Epoch [8/10], Step [800/3000], Loss: 0.1059, Accuracy: 100.00% Epoch [8/10], Step [900/3000], Loss: 0.1013, Accuracy: 100.00% Epoch [8/10], Step [1000/3000], Loss: 0.1359, Accuracy: 100.00% Epoch [8/10], Step [1100/3000], Loss: 0.0577, Accuracy: 100.00% Epoch [8/10], Step [1200/3000], Loss: 0.0268, Accuracy: 100.00% Epoch [8/10], Step [1300/3000], Loss: 0.0948, Accuracy: 95.00% Epoch [8/10], Step [1400/3000], Loss: 0.3132, Accuracy: 95.00% Epoch [8/10], Step [1500/3000], Loss: 0.2723, Accuracy: 85.00% Epoch [8/10], Step [1600/3000], Loss: 0.1014, Accuracy: 100.00% Epoch [8/10], Step [1700/3000], Loss: 0.1055, Accuracy: 100.00% Epoch [8/10], Step [1800/3000], Loss: 0.2086, Accuracy: 90.00% Epoch [8/10], Step [1900/3000], Loss: 0.1960, Accuracy: 95.00% Epoch [8/10], Step [2000/3000], Loss: 0.2814, Accuracy: 90.00% Epoch [8/10], Step [2100/3000], Loss: 0.0181, Accuracy: 100.00% Epoch [8/10], Step [2200/3000], Loss: 0.2399, Accuracy: 85.00% Epoch [8/10], Step [2300/3000], Loss: 0.0969, Accuracy: 95.00% Epoch [8/10], Step [2400/3000], Loss: 0.3500, Accuracy: 85.00% Epoch [8/10], Step [2500/3000], Loss: 0.2651, Accuracy: 90.00% Epoch [8/10], Step [2600/3000], Loss: 0.3941, Accuracy: 95.00% Epoch [8/10], Step [2700/3000], Loss: 0.1631, Accuracy: 95.00% Epoch [8/10], Step [2800/3000], Loss: 0.1636, Accuracy: 90.00% Epoch [8/10], Step [2900/3000], Loss: 0.0246, Accuracy: 100.00% Epoch [8/10], Step [3000/3000], Loss: 0.7887, Accuracy: 90.00% Epoch [9/10], Step [100/3000], Loss: 0.1265, Accuracy: 95.00% Epoch [9/10], Step [200/3000], Loss: 0.0516, Accuracy: 100.00% Epoch [9/10], Step [300/3000], Loss: 0.1836, Accuracy: 95.00% Epoch [9/10], Step [400/3000], Loss: 0.3616, Accuracy: 90.00% Epoch [9/10], Step [500/3000], Loss: 0.2739, Accuracy: 90.00% Epoch [9/10], Step [600/3000], Loss: 0.0896, Accuracy: 95.00% Epoch [9/10], Step [700/3000], Loss: 0.1399, Accuracy: 95.00% Epoch [9/10], Step [800/3000], Loss: 0.2549, Accuracy: 90.00% Epoch [9/10], Step [900/3000], Loss: 0.0927, Accuracy: 95.00% Epoch [9/10], Step [1000/3000], Loss: 0.1919, Accuracy: 95.00% Epoch [9/10], Step [1100/3000], Loss: 0.4908, Accuracy: 85.00% Epoch [9/10], Step [1200/3000], Loss: 0.3770, Accuracy: 95.00% Epoch [9/10], Step [1300/3000], Loss: 0.0630, Accuracy: 100.00% Epoch [9/10], Step [1400/3000], Loss: 0.2617, Accuracy: 85.00% Epoch [9/10], Step [1500/3000], Loss: 0.2551, Accuracy: 90.00% Epoch [9/10], Step [1600/3000], Loss: 0.4593, Accuracy: 85.00% Epoch [9/10], Step [1700/3000], Loss: 0.0568, Accuracy: 100.00% Epoch [9/10], Step [1800/3000], Loss: 0.1691, Accuracy: 95.00% Epoch [9/10], Step [1900/3000], Loss: 0.2033, Accuracy: 95.00% Epoch [9/10], Step [2000/3000], Loss: 0.1835, Accuracy: 90.00% Epoch [9/10], Step [2100/3000], Loss: 0.3293, Accuracy: 90.00% Epoch [9/10], Step [2200/3000], Loss: 0.0377, Accuracy: 100.00% Epoch [9/10], Step [2300/3000], Loss: 0.1097, Accuracy: 100.00% Epoch [9/10], Step [2400/3000], Loss: 0.0347, Accuracy: 100.00% Epoch [9/10], Step [2500/3000], Loss: 0.0739, Accuracy: 95.00% Epoch [9/10], Step [2600/3000], Loss: 0.1824, Accuracy: 90.00% Epoch [9/10], Step [2700/3000], Loss: 0.0137, Accuracy: 100.00% Epoch [9/10], Step [2800/3000], Loss: 0.0946, Accuracy: 100.00% Epoch [9/10], Step [2900/3000], Loss: 0.3189, Accuracy: 90.00% Epoch [9/10], Step [3000/3000], Loss: 0.1116, Accuracy: 90.00% Epoch [10/10], Step [100/3000], Loss: 0.0453, Accuracy: 100.00% Epoch [10/10], Step [200/3000], Loss: 0.0615, Accuracy: 100.00% Epoch [10/10], Step [300/3000], Loss: 0.1178, Accuracy: 95.00% Epoch [10/10], Step [400/3000], Loss: 0.1984, Accuracy: 90.00% Epoch [10/10], Step [500/3000], Loss: 0.1226, Accuracy: 95.00% Epoch [10/10], Step [600/3000], Loss: 0.1297, Accuracy: 95.00% Epoch [10/10], Step [700/3000], Loss: 0.0971, Accuracy: 95.00% Epoch [10/10], Step [800/3000], Loss: 0.1809, Accuracy: 95.00% Epoch [10/10], Step [900/3000], Loss: 0.2740, Accuracy: 95.00% Epoch [10/10], Step [1000/3000], Loss: 0.2821, Accuracy: 85.00% Epoch [10/10], Step [1100/3000], Loss: 0.0506, Accuracy: 100.00% Epoch [10/10], Step [1200/3000], Loss: 0.1346, Accuracy: 95.00% Epoch [10/10], Step [1300/3000], Loss: 0.1816, Accuracy: 90.00% Epoch [10/10], Step [1400/3000], Loss: 0.2985, Accuracy: 95.00% Epoch [10/10], Step [1500/3000], Loss: 0.0861, Accuracy: 100.00% Epoch [10/10], Step [1600/3000], Loss: 0.0410, Accuracy: 100.00% Epoch [10/10], Step [1700/3000], Loss: 0.1141, Accuracy: 95.00% Epoch [10/10], Step [1800/3000], Loss: 0.3941, Accuracy: 90.00% Epoch [10/10], Step [1900/3000], Loss: 0.0578, Accuracy: 100.00% Epoch [10/10], Step [2000/3000], Loss: 0.1544, Accuracy: 95.00% Epoch [10/10], Step [2100/3000], Loss: 0.1653, Accuracy: 90.00% Epoch [10/10], Step [2200/3000], Loss: 0.4368, Accuracy: 95.00% Epoch [10/10], Step [2300/3000], Loss: 0.1606, Accuracy: 90.00% Epoch [10/10], Step [2400/3000], Loss: 0.2422, Accuracy: 90.00% Epoch [10/10], Step [2500/3000], Loss: 0.1275, Accuracy: 95.00% Epoch [10/10], Step [2600/3000], Loss: 0.1298, Accuracy: 95.00% Epoch [10/10], Step [2700/3000], Loss: 0.1009, Accuracy: 100.00% Epoch [10/10], Step [2800/3000], Loss: 0.1852, Accuracy: 90.00% Epoch [10/10], Step [2900/3000], Loss: 0.0445, Accuracy: 100.00% Epoch [10/10], Step [3000/3000], Loss: 0.3749, Accuracy: 85.00%
PLOT ACCURACY AND LOSS for each iteration
In [64]:
In [63]:
PLOT average ACCURACY AND LOSS for each epoc
In [71]:
Out[71]:
(0.8, 1.0)
In [72]:
Out[72]:
(0.0, 0.2)
In [0]:
Out[0]:
<torch.utils.data.dataloader.DataLoader at 0x7f9133b177f0>
In [0]:
Test Accuracy of the model on the 10000 test images: 91.34 %
Compute a per class accuracy of your classifier on the validation and test dataset
In [0]:
Accuracy of T-shirt/top : 81 % Accuracy of Trouser : 99 % Accuracy of Pullover : 83 % Accuracy of Dress : 91 % Accuracy of Coat : 90 % Accuracy of Sandal : 97 % Accuracy of Shirt : 77 % Accuracy of Sneaker : 97 % Accuracy of Bag : 98 % Accuracy of Ankle boot : 95 %
Which classes are the hardest to get? Show 2 images from each class which the network classifies correctly, and 2 more images where it classifies incorrectly. I am showing images for each category, but the hardest were t-shirts and shirts which make sense since these can easily look like other things (more ambiguous, less definite features).
In [0]:
tensor(9)_true Ankle boot
tensor(2)_true Pullover
tensor(1)_true Trouser
tensor(6)_true Shirt
tensor(4)_true Coat
tensor(5)_true Sandal
tensor(7)_true Sneaker
tensor(3)_true Dress
tensor(8)_true Bag
tensor(0)_true T-shirt/top
tensor(5)_false Sandal
tensor(4)_false Coat
tensor(3)_false Dress
tensor(6)_false Shirt
tensor(2)_false Pullover
tensor(9)_false Ankle boot
tensor(0)_false T-shirt/top
tensor(8)_false Bag
tensor(1)_false Trouser
tensor(7)_false Sneaker
FILTERS
In [0]:
(32, 1, 3, 3) First convolutional layer
Second convolutional layer
In [0]:
Out[0]:
(32, 32, 3, 3)