search

tallamjr / astronet

Efficient Deep Learning for Real-time Classification of Astronomical Transients and Multivariate Time-series
Apache License 2.0
14 stars 3 forks source link

Custom loss function error: TypeError: __init__() takes from 1 to 2 positional arguments but 3 were given #54

Closed tallamjr closed 3 years ago

tallamjr commented 3 years ago

This is well summarised here https://github.com/tensorflow/tensorflow/issues/45079 where someone is trying to do essentially the same as is done in astronet i.e. define a custom loss function and use this to optimise for.

When running a test function such as:

$ python astronet/t2/train.py --dataset "plasticc" --epoch 1 --batch-size 256

the follow error occurs from commit 403fcdde7121739239bdaac166426c405fa883f6 onwards (discoverd with git bisect and the test defined above)

(astronet) 15:19:52 ✔ ~/github/tallamjr/origin/astronet (issues/53/load-datasets) :: python astronet/t2/train.py --dataset "plasticc" --epoch 1 --batch-size 256
[20-12-18 15:20:15] {train.py:28} INFO - =======================================================================================================================================
[20-12-18 15:20:15] {train.py:29} INFO - File Path: /Users/tallamjr/github/tallamjr/origin/astronet/astronet/t2/train.py
[20-12-18 15:20:15] {train.py:30} INFO - Parent of Directory Path: /Users/tallamjr/github/tallamjr/origin
(2991, 100, 6) (2991, 3)
[20-12-18 15:20:17] {train.py:54} INFO - None
(256, 100, 6)
Traceback (most recent call last):
  File "astronet/t2/train.py", line 191, in <module>
    training()
  File "astronet/t2/train.py", line 126, in __call__
    mode="min",
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 1103, in fit
    tmp_logs = self.train_function(iterator)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 808, in train_function
    return step_function(self, iterator)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 798, in step_function
    outputs = model.distribute_strategy.run(run_step, args=(data,))
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/distribute/distribute_lib.py", line 1259, in run
    return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/distribute/distribute_lib.py", line 2731, in call_for_each_replica
    return self._call_for_each_replica(fn, args, kwargs)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/distribute/distribute_lib.py", line 3420, in _call_for_each_replica
    return fn(*args, **kwargs)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 572, in wrapper
    return func(*args, **kwargs)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 791, in run_step
    outputs = model.train_step(data)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 759, in train_step
    y, y_pred, sample_weight, regularization_losses=self.losses)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/engine/compile_utils.py", line 204, in __call__
    loss_value = loss_obj(y_t, y_p, sample_weight=sw)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/losses.py", line 152, in __call__
    losses = call_fn(y_true, y_pred)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/keras/losses.py", line 256, in call
    return ag_fn(y_true, y_pred, **self._fn_kwargs)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 667, in wrapper
    return converted_call(f, args, kwargs, options=options)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 393, in converted_call
    return _call_unconverted(f, args, kwargs, options)
  File "/usr/local/anaconda3/envs/astronet/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 478, in _call_unconverted
    return f(*args, **kwargs)
TypeError: __init__() takes from 1 to 2 positional arguments but 3 were given

As this is so closely related to the tensorflow issue linked above and many avenues exhausted, it is felt best to hold off until there is further updates in the thread of a solution is posted.

tallamjr commented 3 years ago

This has potentially been fixed with:

-        for train_index, val_index in skf.split(X_train, y_train.argmax(1)):
+
+        print(type(y_train))
+        if tf.is_tensor(y_train):
+            y_train = np.array(y_train)
+            print(type(y_train))
+            y_train_split = y_train.argmax(1)
+            print(y_train_split)
+        else:
+            y_train_split = y_train.argmax(1)
+            print(y_train_split)
+
+        for train_index, val_index in skf.split(X_train, y_train_split):
             X_train_cv, X_val_cv = X_train[train_index], X_train[val_index]
             y_train_cv, y_val_cv = y_train[train_index], y_train[val_index]

When #56 is closed, then this can be properly tested with a new PLAsTiCC analysis being able to be run

tallamjr commented 3 years ago

Closed with #59