my.pytorch's Introduction

my.pytorch

My codebase to train and test various networks

Mnist

Resnet
Mnist CNN
FC

Appendix

Leaderboard

Provide leaderboard with top1 config and model symlink

#Rank : Data                          ,   Test Acc   Test Loss   Train Acc  Train Loss
   1  : ../Mar09-2054/final.pth       ,   0.993700,   0.025870,   0.999500,   0.001450
   2  : ../Mar09-2025/final.pth       ,   0.993300,   0.024709,   0.999800,   0.000804
   3  : ../Mar09-1630/final.pth       ,   0.993200,   0.023206,   1.000000,   0.001310
   ...

Includable YAML

# configs/mnist_cnn.yaml

# MNIST dataset
DATASET: !include dataset/mnist.yaml
MODEL: !include model/mnist_base.yaml

TRAIN:
  EPOCH: 50
  BATCH_SIZE: 512
  OPTIMIZER:
    PARAM:
      lr: 0.05
      momentum: 0.9
      weight_decay: 0.0001
  LR_SCHEDULER: !include lr_scheduler/sgdr.yaml

my.pytorch's People

Contributors

Watchers

my.pytorch's Issues

Implement inference

mAP summary per classes

As well as top1-error, top5-error, logging

onnx exporter

Implement checkpoint load

Re-enable back the Non-existent config key check in attrdict


-        if k not in b:
-            raise KeyError('Non-existent config key: {}'.format(full_key))
+        #TODO(jihoonl): This is good sanity check. Please re-enable back
+        #if k not in b:
+        #    raise KeyError('Non-existent config key: {}'.format(full_key))

         v = _decode_cfg_value(v_)
-        v = _check_and_coerce_cfg_value_type(v, b[k], k,
-                                             full_key)  # TODO(jihoonl): WTF
+        #v = _check_and_coerce_cfg_value_type(v, b[k], k,
+        #                                     full_key)  # TODO(jihoonl): WTF