When running,I have an error

[gaojunhui68]

When running,I have an error:

Traceback (most recent call last):
File "example.py", line 9, in
model.train(ckpt_dir='ckpt')
File "D:\mypy\ORGAN-master\organ_init_.py", line 763, in train
gen_samples, self.train_samples, self.ord_dict, results)
File "D:\mypy\ORGAN-master\organ\mol_metrics.py", line 185, in compute_results
results[objective] = np.mean(reward(verified_samples, train_data))
File "D:\mypy\ORGAN-master\organ_init_.py", line 743, in batch_reward
for sample in samples]
File "D:\mypy\ORGAN-master\organ_init_.py", line 743, in
for sample in samples]
File "D:\mypy\ORGAN-master\organ\mol_metrics.py", line 117, in decode
''.join([ord_dict[o] for o in ords]))
File "D:\mypy\ORGAN-master\organ\mol_metrics.py", line 117, in
''.join([ord_dict[o] for o in ords]))
KeyError: 'O'

[gaojunhui68]

Traceback (most recent call last):
File "example.py", line 18, in
model.train() # Proceeds with the training
File "D:\mypy\ORGAN-master\organ_init_.py", line 763, in train
gen_samples, self.train_samples, self.ord_dict, results)
File "D:\mypy\ORGAN-master\organ\music_metrics.py", line 252, in compute_results
results[key] = np.mean(reward(samples, train_samples))
File "D:\mypy\ORGAN-master\organ_init_.py", line 743, in batch_reward
for sample in samples]
File "D:\mypy\ORGAN-master\organ_init_.py", line 743, in
for sample in samples]
File "D:\mypy\ORGAN-master\organ\music_metrics.py", line 60, in decode
return ' '.join(unpad([ord_dict[o] for o in ords]))
File "D:\mypy\ORGAN-master\organ\music_metrics.py", line 60, in
return ' '.join(unpad([ord_dict[o] for o in ords]))
KeyError: 'b'

[couteiral]

Hi @gaojunhui68,

Both errors come (obviously) from different sources. Particularly, the first is using the molecular metrics, and the second is using the music metrics.

In both cases, it looks like you are using different training sets in the run and the checkpoint, so the engine is unable to decode (because the internal dictionary does not recognize the features. If you give me more information (i. e., the actual file that you run), I'll be able to give you more information.

Regards,
Carlos

[gaojunhui68]

Hi @couteiral,

Yes, the first is using the molecular metrics, and the second is using the music metrics.

For the first, the code of example.py is in bellow:

import organ
from organ import ORGAN
model = ORGAN('test', 'mol_metrics', params={'PRETRAIN_DIS_EPOCHS': 1})
model.load_training_set('data/toy.csv')
model.set_training_program(['novelty'], [1])
model.load_metrics()
model.train(ckpt_dir='ckpt')

For the second , the code of example.py is in bellow:

from organ import ORGAN

model = ORGAN('test', 'music_metrics') # Loads a ORGANIC with name 'test', using music metrics
model.load_training_set('data/music_small.txt') # Loads the training set
model.set_training_program(['tonality'], [50]) # Sets the training program as 50 epochs with the tonality metric
model.load_metrics() # Loads all the metrics
model.train() # Proceeds with the training

In both cases ,this error occurs in train, after pretrain.

Please help me .
Thanks,
Junhui Gao

[couteiral]

Hi @gaojunhui68,

First, the music metrics seem to be bugged. I am afraid I didn't work on them myself, but I'll get in touch with someone involved, and get back to you.

Regarding the molecular metrics, you get a KeyError, which is the error that a Python dictionary raises when a key not in the dictionary is requested. The following is happening: when you try to decode the embedding coordinates to SMILES strings, you are passing the wrong value to the dictionary, and the code crashes.

In particular, you are passing 'O' to the ord_dict, which is the dictionary containing the mapping from the embedding to the SMILES strings, so something is wrong in there. However, I just ran exactly the same code from the actual repo, and I could not found any problem like yours.

Could you share your pretraining files, so I can have a look at them? Also, are you sure that there is nothing wrong with your 'toy.csv' training set?

Cheers,
Carlos

[n-yoshikawa]

Hi @couteiral,

I had the same error as @gaojunhui68. What I did was:

git clone https://github.com/gablg1/ORGAN.git
pip install -r requirements.txt
python example.py

The error message was

Traceback (most recent call last):
  File "example.py", line 8, in <module>
    model.train(ckpt_dir='ckpt')
  File "/home/yoshikawa/ORGAN/organ/__init__.py", line 763, in train
    gen_samples, self.train_samples, self.ord_dict, results)
  File "/home/yoshikawa/ORGAN/organ/mol_metrics.py", line 183, in compute_results
    results[objective] = np.mean(reward(verified_samples, train_data))
  File "/home/yoshikawa/ORGAN/organ/__init__.py", line 743, in batch_reward
    for sample in samples]
  File "/home/yoshikawa/ORGAN/organ/mol_metrics.py", line 115, in decode
    ''.join([ord_dict[o] for o in ords]))
KeyError: 'O'

I used pyenv. Both anaconda3-5.0.0 and anaconda2-5.0.0 did not work.

[toushi68]

Hi @couteiral,

Any update for the previous posts? The error message I got is KeyError: 'N'

I was also running the example.py. It happened at "model.train(ckpt_dir='ckpt')", after finished the pre-training. It looks like it was happened at the same spot like the KeyError: 'O'.

Traceback (most recent call last):
File "", line 1, in
File "/home/trial0/ORGAN/organ/init.py", line 763, in train
gen_samples, self.train_samples, self.ord_dict, results)
File "/home/trial0/ORGAN/organ/mol_metrics.py", line 183, in compute_results
results[objective] = np.mean(reward(verified_samples, train_data))
File "/home/trial0/ORGAN/organ/init.py", line 743, in batch_reward
for sample in samples]
File "/home/trial0/ORGAN/organ/init.py", line 743, in
for sample in samples]
File "/home/trial0/ORGAN/organ/mol_metrics.py", line 115, in decode
''.join([ord_dict[o] for o in ords]))
File "/home/trial0/ORGAN/organ/mol_metrics.py", line 115, in
''.join([ord_dict[o] for o in ords]))
KeyError: 'N'

Is there anyway to pin point which smiles made the problem?
Please let me know if you need any further details.

Thanks in advance!
Toushi

[toushi68]

It looks like this issue is related to the data set. For the toy set, I found there are > 30 entries with empty smiles, i.e. have NumAtom, Name, but the smiles column are empty. From there I further refined the data set with rdkit. With all these trials, I got different KeyError(s), 'C', '[', 'O'. This means the data set still has something wrong! Or a filter is needed before processing the data just like "ORGANIC" does.

[yippp]

Hi @couteiral,
I found that the error is due to the function mm.decode(ords, rod_dict). The ords may be string or list. Thats why it will occurs an error.
I think that there should be 2 different decode() functions.

Could you run the code use your music_small.txt dataset again to help to fix it?
Thanks.

[toushi68]

Hi @couterial,

Have a look of some simple debug.
If I insert a print in the decode, like this:

def decode(ords, ord_dict):
print (ords) # check
return unpad(''.join([ord_dict[o] for o in ords]))

Here are the last few lines printed out before it crashes:
.........
[11 1 2 1 1 2 1 1 2 1 4 11 7 21 21 21]
[ 1 2 11 1 1 9 11 10 2 11 21 21 21 21 21 21]
[ 1 2 1 1 9 8 10 8 21 21 21 21 21 21 21 21]
[11 2 1 1 1 3 4 3 9 7 8 10 6 4 21 21]
[ 8 1 7 3 4 5 5 3 5 6 4 21 21 21 21 21]
N#CC(O)F
Traceback (most recent call last):
File "/home/trial3/ORGAN/organ/init.py", line 763, in train
gen_samples, self.train_samples, self.ord_dict, results)
File "/home/trial3/ORGAN/organ/mol_metrics.py", line 185, in compute_results
results[objective] = np.mean(reward(verified_samples, train_data))
File "/home/trial3/ORGAN/organ/init.py", line 743, in batch_reward
for sample in samples]
File "/home/trial3/ORGAN/organ/mol_metrics.py", line 117, in decode
return unpad(''.join([ord_dict[o] for o in ords]))
KeyError: 'N'

It looks like this is an already decoded smile, which should not be sent back to decode again.
Any idea what's going on?
Thanks!
Toushi68

[k105la]

Hello @couteiral

I ran my code line by line to see where the problem occurs. I see that after importing the dataset with model.load_training_set('data/toy.csv') the error comes up.

Traceback (most recent call last):
File "", line 1, in
File "/Users/akilhylton/Desktop/ORGAN/organ/init.py", line 242, in load_training_set
self.char_dict) for sam in to_use]
File "/Users/akilhylton/Desktop/ORGAN/organ/init.py", line 242, in
self.char_dict) for sam in to_use]
File "/Users/akilhylton/Desktop/ORGAN/organ/mol_metrics.py", line 383, in encode
return [char_dict[c] for c in pad(new_smi, max_len)]
File "/Users/akilhylton/Desktop/ORGAN/organ/mol_metrics.py", line 383, in
return [char_dict[c] for c in pad(new_smi, max_len)]
KeyError: '.'

Any solutions?

[Kajiyu]

I have the same error with @ahylton19 .

Traceback (most recent call last):
File "example.py", line 5, in
model.load_training_set('data/toy.csv')
File "/home/yuma_kajihara/projects/ORGAN/organ/init.py", line 242, in load_training_set
self.char_dict) for sam in to_use]
File "/home/yuma_kajihara/projects/ORGAN/organ/init.py", line 242, in
self.char_dict) for sam in to_use]
File "/home/yuma_kajihara/projects/ORGAN/organ/mol_metrics.py", line 384, in encode
return [char_dict[c] for c in pad(new_smi, max_len)]
File "/home/yuma_kajihara/projects/ORGAN/organ/mol_metrics.py", line 384, in
return [char_dict[c] for c in pad(new_smi, max_len)]
KeyError: '.'

[xuzhang5788]

I have the same error with @Kajiyu
Traceback (most recent call last):
File "example.py", line 5, in
model.load_training_set('data/toy.csv')
File "/media/projects/ORGAN/organ/init.py", line 242, in load_training_set
self.char_dict) for sam in to_use]
File "/media/projects/ORGAN/organ/init.py", line 242, in
self.char_dict) for sam in to_use]
File "/media/projects/ORGAN/organ/mol_metrics.py", line 383, in encode
return [char_dict[c] for c in pad(new_smi, max_len)]
File "/media/projects/ORGAN/organ/mol_metrics.py", line 383, in
return [char_dict[c] for c in pad(new_smi, max_len)]
KeyError: '.'

Any solutions?

[beangoben]

we will be updating the repo soon, we expect these changes to be incorporated by monday, latest tuesday...stay tuned, they will fix these issues.

[kristery]

I have the same error with @xuzhang5788
Seems they are not going to update it?

[kristery]

I checked the table of SMILES and it seems that '.' represents one kind of bond and in the mol_metrics.py file they didn't add it. To fix the error please modify line 315 to

chars = chars + ['-', '=', '#', '.']

I guess it works as long as you add '.' to chars.