Centralize logging for uniform CLI output

[tonyzamyatin]

Description:

Currently, additional logging outside of WandB is handled via dispersed print statements throughout the code, which do not follow a uniform format. This results in inconsistent and sometimes cluttered CLI output, for example:

INFO: Precomputing 13941 items...
[19:27:41] Conflicting single bond directions around double bond at index 7.
[19:27:41]   BondStereo set to STEREONONE and single bond directions set to NONE.
[19:27:41] Conflicting single bond directions around double bond at index 7.
[19:27:41]   BondStereo set to STEREONONE and single bond directions set to NONE.
INFO: Precomputation finished in 51.36s.
INFO: Precomputing 774 items...
INFO: Precomputation finished in 2.84s.
INFO: Precomputing 775 items...
INFO: Precomputation finished in 2.90s.
INFO: Updating global config with properties of training dataset
INFO: Final config:
data_ingestor:
  data_source:
    _target_: chemtorch.data_ingestor.data_source.SingleCSVSource
    data_path: data/rdb7/barriers/forward/data.csv
    ...
dataset:
  _target_: chemtorch.dataset.GraphDataset
  ...
  representation:
    _target_: chemtorch.representation.graph.cgr.CGR
    ...
dataloader:
  _target_: torch_geometric.loader.DataLoader
  batch_size: 50
  ...
model:
  ...
routine:
  ...
...
Total parameters: 305,921
GNN(
  (encoder): DirectedEdgeEncoder(
    (edge_init): Linear(in_features=110, out_features=128, bias=True)
  )
  (layer_stack): DMPNNStack(
    (dmpnn_blocks): LayerStack(
      (layers): ModuleList(
        (0-2): 3 x DMPNNBlock(
          (graph_conv): DMPNNConv()
          (activation): ReLU()
          (norm): Identity()
          (dropout): Dropout(p=0.1, inplace=False)
          (ffn_norm_in): Identity()
          (ffn_linear1): Linear(in_features=128, out_features=256, bias=True)
          (ffn_linear2): Linear(in_features=256, out_features=128, bias=True)
          (ffn_act_fn): ReLU()
          (ffn_norm_out): Identity()
          (ffn_dropout1): Dropout(p=0.1, inplace=False)
          (ffn_dropout2): Dropout(p=0.1, inplace=False)
        )
      )
    )
    (edge_to_node_embedding): EdgeToNodeEmbedding(
      (linear): Linear(in_features=216, out_features=128, bias=True)
      (activation): ReLU()
      (aggregation): SumAggregation()
    )
  )
  (pool): GlobalPool()
  (head): MLP(
    (activation): ReLU()
    (layers): Sequential(
      (0): Dropout(p=0.02, inplace=False)
      (1): Linear(in_features=128, out_features=128, bias=True)
      (2): ReLU()
      (3): Dropout(p=0.02, inplace=False)
      (4): Linear(in_features=128, out_features=1, bias=True)
    )
  )
)

Proposal:
Implement a centralized, multilevel logging system (e.g., using Python’s built-in logging module) to:

• Replace print statements with proper logging calls (e.g., logger.info, logger.warning, logger.error, etc.).
• Unify log formatting for all output, making CLI output consistent and easy to follow.
• Support multiple log levels (DEBUG, INFO, WARNING, ERROR).
• Make future maintenance easier and improve developer/user experience.