Coursework on data processing.
Convolutional neural networks with 3 and 4 parallel input layers are constructed, processing the trigram form of sentences.
| 1 | 2 | 3 |
|---|---|---|
| create_dataset.ipynb | dataset_preprocessing.ipynb | analysys.ipynb |
 |
|
|
The scripts are executed completely sequentially. The first two do not need computers more powerful than the average workforce (laptop, for example).
However, the * analysys.ipynb * block requires quite powerful machines, the ** GPU ** accelerator is recommended (the rendering was performed on Kaggle, where the training time was already acceptable with the enabled GPU).
- 4 grams. Still, the results, as shown by the applications before, will be better;
- Increasing the volume of the training sample. Increases the cost of computing power.