Failed to run PS3E25 LightAutoML Neural Nets baseline in VS Code #145
Comments
|
@sandeshregmi please try to install LightAutoML before importing any library, that should help. |
|
@alexmryzhkov Thank you for your suggestion. I followed your suggestion but I still got same error. Here are the list of package dependencies that installed before I imported other packages. |
|
Hello @sandeshregmi! Can you please tell me the way you installed LightAutoML? I have tried to run the code with the installed Moreover, I have tried the same with cloning and building the LightAutoML master branch and, again, have successfully run the scripts in the notebook. Nevertheless, I was able to reproduce this error by installing LightAutoML with If you used |
|
@BELONOVSKII Thank you! for me following version worked |
|
@BELONOVSKII Also, do you have a suggestion or (notebook) to do model explanation analysis for a model built by for example 'fttransformer'? |
|
@sandeshregmi you can use this kernel as a reference how to use FTTransformer - it is the last one in the list. |
|
@alexmryzhkov Thank you! Sorry I was not very descriptive in my question. Do you think we can generate something similar to SHAP plot ? Or Attention weights based feature importance? |
|
@sandeshregmi Hello! LightAutoML has an implementation of SSWARM algorithm for model-agnostic computations of Shapley values. This algorithm computes Shapley values for compositions of any ML models (not only tree-based). For now, this algorithm is not released in GitHub or PyPI (will come in the near future), nevertheless, I have a |
|
@BELONOVSKII Thank you! |
|
@sandeshregmi yes, you need to uninstall 0.3.9b1 and use the whl file @BELONOVSKII provided because it is the only current version which hav SSWARM algo to calc shapely values based on lightautoml models |
|
@BELONOVSKII Awesome! It worked. Thank you |
|
@BELONOVSKII One thing I noticed is though |
|
@sandeshregmi Hello. |
|
@BELONOVSKII Could you please review this code. I am getting same feature name for
|
|
@sandeshregmi Hello! Yes, in this code I have manually set the interaction feature for each plot ( the first feature in the list ). However, you can set any other feature for interaction, or set the parameter to Read more in the |
I downloaded .ipynb file from kaggle and tried running in VS code. It worked for LGBM algorithm but not others. Here is an error I got when I ran MLP.
[16:30:57] Stdout logging level is INFO3.
[16:30:57] Task: reg
[16:30:57] Start automl preset with listed constraints:
[16:30:57] - time: 10000.00 seconds
[16:30:57] - CPU: 4 cores
[16:30:57] - memory: 16 GB
[16:30:57] Train data shape: (11651, 12)
[16:30:57] Layer 1 train process start. Time left 9999.99 secs
[16:30:57] Start fitting Lvl_0_Pipe_0_Mod_0_TorchNN_mlp_0 ...
[16:30:57] ===== Start working with fold 0 for Lvl_0_Pipe_0_Mod_0_TorchNN_mlp_0 =====
[16:30:57] Model Lvl_0_Pipe_0_Mod_0_TorchNN_mlp_0 failed during ml_algo.fit_predict call.
'str' object is not callable
[16:30:57] Traceback (most recent call last):
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\ml_algo\utils.py", line 72, in tune_and_fit_predict
preds = ml_algo.fit_predict(train_valid)
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\ml_algo\dl_model.py", line 472, in fit_predict
return super().fit_predict(train_valid_iterator)
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\ml_algo\base.py", line 273, in fit_predict
model, pred = self.fit_predict_single_fold(train, valid)
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\ml_algo\dl_model.py", line 498, in fit_predict_single_fold
val_pred = model.fit(dataloaders)
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\text\trainer.py", line 430, in fit
self._init()
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\text\trainer.py", line 335, in _init
self.model = self.net(**self.net_params)
File "c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\text\nn_model.py", line 295, in init
self.cont_embedder = cont_embedder(**cont_params)
TypeError: 'str' object is not callable
AssertionError Traceback (most recent call last)
Cell In[27], line 20
1 algo = 'mlp'
2 automl = TabularAutoML(
3 task = task,
4 timeout = TIMEOUT,
(...)
17 reader_params = {'n_jobs': N_THREADS, 'cv': N_FOLDS, 'random_state': RANDOM_STATE, 'advanced_roles': ADVANCED_ROLES}
18 )
---> 20 oof_pred = automl.fit_predict(tr_data, roles = roles, verbose = 3)
21 test_pred = automl.predict(te_data)
22 oof = score(task, mapped(tr_data[TARGET_NAME].values, task, automl.reader.class_mapping), take_pred_from_task(oof_pred.data, task))
File c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\automl\presets\tabular_presets.py:709, in TabularAutoML.fit_predict(self, train_data, roles, train_features, cv_iter, valid_data, valid_features, log_file, verbose)
706 if self.is_time_series:
707 train = {"seq": {"seq0": train}}
--> 709 oof_pred = super().fit_predict(train, roles=roles, cv_iter=cv_iter, valid_data=valid_data, verbose=verbose)
711 return cast(NumpyDataset, oof_pred)
File c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\automl\presets\base.py:207, in AutoMLPreset.fit_predict(self, train_data, roles, train_features, cv_iter, valid_data, valid_features, verbose)
204 logger.info(f"- memory: {self.memory_limit} GB\n")
206 self.timer.start()
--> 207 result = super().fit_predict(
208 train_data,
209 roles,
210 train_features,
211 cv_iter,
212 valid_data,
213 valid_features,
214 verbose=verbose,
215 )
217 logger.info("\x1b[1mAutoml preset training completed in {:.2f} seconds\x1b[0m\n".format(self.timer.time_spent))
218 logger.info(f"Model description:\n{self.create_model_str_desc()}\n")
File c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\automl\base.py:218, in AutoML.fit_predict(self, train_data, roles, train_features, cv_iter, valid_data, valid_features, verbose)
216 for k, ml_pipe in enumerate(level):
217 ml_pipe.debug = self.debug
--> 218 pipe_pred = ml_pipe.fit_predict(train_valid)
219 level_predictions.append(pipe_pred)
220 pipes.append(ml_pipe)
File c:\Users\sregmi444\Anaconda3\envs\test\lib\site-packages\lightautoml\pipelines\ml\base.py:135, in MLPipeline.fit_predict(self, train_valid)
131 self.ml_algos.append(ml_algo)
133 predictions.append(preds)
--> 135 assert (
136 len(predictions) > 0
137 ), "Pipeline finished with 0 models for some reason.\nProbably one or more models failed"
139 predictions = concatenate(predictions)
141 del self._ml_algos
AssertionError: Pipeline finished with 0 models for some reason.
Probably one or more models failed
The text was updated successfully, but these errors were encountered: