feat: update SYMBA_HEP and SYMBA_REG implementations

SYMBA_HEP/SYMBAHEP_Hybrid_SSM_Prasanth_Naidu/model/helpers/mamba.py

@@ -239,7 +239,6 @@ def forward(
         hidden_states = self.embedding.forward(input_ids)
         residual = None
         for idx, layer in enumerate(self.layers):
-            # print('layer: ',idx)
             hidden_states, residual = layer(
                 hidden_states,
                 context=context,
@@ -295,12 +294,10 @@ def __init__(
         factory_kwargs = {"device": device, "dtype": dtype}
 
         super().__init__()
-        # print(vocab_size)
         # if vocab_size % pad_vocab_size_multiple != 0:
         #     vocab_size += pad_vocab_size_multiple - (
         #         vocab_size % pad_vocab_size_multiple
         #     )
-        # print(vocab_size)
 
         self.backbone = MixerModel(
             d_model=d_model,

SYMBA_HEP/SYMBAHEP_Hybrid_SSM_Prasanth_Naidu/model/mamba_encdec.py

@@ -138,7 +138,6 @@ def forward(
             input_ids=context_tokens,
             mask=source_attention_mask,
         )
-        # print(source_vec.dtype, source_attention_mask.dtype)
         cache = self.allocate_inference_cache(
             batch_size=b,
             max_seqlen=300 + l + 1,  # source + BOS
@@ -153,10 +152,6 @@ def forward(
         # batch, seqlen, dim = self.decoder.backbone.embedding.forward(input_ids).shape
         # conv_state, ssm_state = self.decoder.backbone.layers[0].mixer._get_states_from_cache(inference_params, b)
         # inference_params = None
-        # print(conv_state.type(),input_ids.type(), source_vec.type())
-        # print(source_attention_mask.type(), target_attention_mask.type())
-        # print(position_ids.type())
-        # print(num_last_tokens)
 
         out = self.decoder.forward(
             input_ids,

SYMBA_HEP/SYMBAHEP_Sym_KAN_TP_Transformer_by_Ayush_Mishra/src/SineKAN_MoE/model.py

@@ -290,7 +290,6 @@ def forward(self , src, tgt):
     dec_output = decod_embed
     for decoder in self.decoder:
       dec_output = decoder(dec_output , enc_output , src_mask , tgt_mask )
-    # print("Decoder output " , dec_output)
     out = self.ff(dec_output)
 
     # out = self.ff(dec_output)

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-08-14-QED-DataPreparation/dev/ampl_raw.py

@@ -59,7 +59,6 @@
         #     break
 
     # sqampls_prefix = []
-    # print("Loading squared amplitudes")
     # for exp in tqdm(sqampl_raw):
     #     # simplified = sp.factor(exp)   # worked best for simplification
     #     # prefix = sympy_to_prefix(simplified)

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-08-14-QED-DataPreparation/dev/implement_hybrid_prefix_and_tokens_combiner.py

@@ -95,22 +95,17 @@ def _queue_mgr(func_str: str, q_in: mp.Queue, q_out: mp.Queue, timeout: int, pid
         proc = mp.Process(target=_lemmiwinks, args=(func_str, (x,), {}, q_worker,))
         proc.start()
         try:
-            # print(f'[{pid}]: {positioning}: getting')
             res = q_worker.get(timeout=timeout)
-            # print(f'[{pid}]: {positioning}: got')
             q_out.put((positioning, res))
         except mpq.Empty:
             q_out.put((positioning, sp.sympify(x)))
-            # print(f'[{pid}]: {positioning}: timed out ({timeout}s)')
             with open(timeout_logfile, "a") as f:
                 f.write("Timed out after "+str(timeout)+" seconds. Argument:" + x + "\n")
         finally:
             try:
                 proc.terminate()
-                # print(f'[{pid}]: {positioning}: terminated')
             except:
                 pass
-    # print(f'[{pid}]: completed!')
 
 
 def killer_pmap(func: Callable, iterable: Iterable, cpus: Optional[int] = None, timeout: int = 10*60,
@@ -140,7 +135,7 @@ def killer_pmap(func: Callable, iterable: Iterable, cpus: Optional[int] = None,
         mp.Process(target=_queue_mgr, args=(dill.dumps(func), q_in, q_out, timeout, pid, timeout_logfile))
         for pid in range(cpus)
     ]
-    # print(f'Started {len(processes)} processes')
+
     for proc in processes:
         proc.start()
 

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-08-14-QED-DataPreparation/scripts/DataPreparation_parallel.py

@@ -165,7 +165,6 @@ def get_unique_indices(l):
     print("Resuming calculations, reading progress from "+progress_file)
     with open(progress_file) as f:
         progess_file_contents = [line for line in f.readlines()]
-    # print(progess_file_contents[-6:])
     batch_resume = int(progess_file_contents[-7].split(":")[1]) + 1
     index_resume = int(progess_file_contents[-3].split(":")[1])
     batch_size_resume = int(progess_file_contents[-2].split(":")[1])

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-08-14-QED-DataPreparation/source/ExpressionsTokensCombiner.py

@@ -13,14 +13,6 @@ def fix_i_match(matchobj):
     """
     match = matchobj.group(0)
     return match.replace("i", "I")
-    # if int(match[1]) % 2 != 0:
-    #     print("asdf")
-    # exponent = int(match[1]) // 2
-    # m, m_name = match[0].split("_")
-    # if exponent == 1:
-    #     return m+"2"+m_name
-    # else:
-    #     return m+"2"+m_name + "**" + str(exponent)
 
 
 def combine_m_match(matchobj):

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-09-15-QCD-DataPreparation/dev/ampl_raw.py

@@ -59,7 +59,6 @@
         #     break
 
     # sqampls_prefix = []
-    # print("Loading squared amplitudes")
     # for exp in tqdm(sqampl_raw):
     #     # simplified = sp.factor(exp)   # worked best for simplification
     #     # prefix = sympy_to_prefix(simplified)

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-09-15-QCD-DataPreparation/dev/implement_hybrid_prefix_and_tokens_combiner.py

@@ -95,22 +95,17 @@ def _queue_mgr(func_str: str, q_in: mp.Queue, q_out: mp.Queue, timeout: int, pid
         proc = mp.Process(target=_lemmiwinks, args=(func_str, (x,), {}, q_worker,))
         proc.start()
         try:
-            # print(f'[{pid}]: {positioning}: getting')
             res = q_worker.get(timeout=timeout)
-            # print(f'[{pid}]: {positioning}: got')
             q_out.put((positioning, res))
         except mpq.Empty:
             q_out.put((positioning, sp.sympify(x)))
-            # print(f'[{pid}]: {positioning}: timed out ({timeout}s)')
             with open(timeout_logfile, "a") as f:
                 f.write("Timed out after "+str(timeout)+" seconds. Argument:" + x + "\n")
         finally:
             try:
                 proc.terminate()
-                # print(f'[{pid}]: {positioning}: terminated')
             except:
                 pass
-    # print(f'[{pid}]: completed!')
 
 
 def killer_pmap(func: Callable, iterable: Iterable, cpus: Optional[int] = None, timeout: int = 10*60,
@@ -140,7 +135,7 @@ def killer_pmap(func: Callable, iterable: Iterable, cpus: Optional[int] = None,
         mp.Process(target=_queue_mgr, args=(dill.dumps(func), q_in, q_out, timeout, pid, timeout_logfile))
         for pid in range(cpus)
     ]
-    # print(f'Started {len(processes)} processes')
+
     for proc in processes:
         proc.start()
 

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-09-15-QCD-DataPreparation/scripts/DataPreparation_parallel.py

@@ -165,7 +165,6 @@ def get_unique_indices(l):
     print("Resuming calculations, reading progress from "+progress_file)
     with open(progress_file) as f:
         progess_file_contents = [line for line in f.readlines()]
-    # print(progess_file_contents[-6:])
     batch_resume = int(progess_file_contents[-7].split(":")[1]) + 1
     index_resume = int(progess_file_contents[-3].split(":")[1])
     batch_size_resume = int(progess_file_contents[-2].split(":")[1])

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-09-15-QCD-DataPreparation/source/ExpressionsTokensCombiner.py

@@ -13,14 +13,6 @@ def fix_i_match(matchobj):
     """
     match = matchobj.group(0)
     return match.replace("i", "I")
-    # if int(match[1]) % 2 != 0:
-    #     print("asdf")
-    # exponent = int(match[1]) // 2
-    # m, m_name = match[0].split("_")
-    # if exponent == 1:
-    #     return m+"2"+m_name
-    # else:
-    #     return m+"2"+m_name + "**" + str(exponent)
 
 
 def combine_m_match(matchobj):

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-10-06-QED-DataPreparation/source/ExpressionsTokensCombiner.py

@@ -13,14 +13,6 @@ def fix_i_match(matchobj):
     """
     match = matchobj.group(0)
     return match.replace("i", "I")
-    # if int(match[1]) % 2 != 0:
-    #     print("asdf")
-    # exponent = int(match[1]) // 2
-    # m, m_name = match[0].split("_")
-    # if exponent == 1:
-    #     return m+"2"+m_name
-    # else:
-    #     return m+"2"+m_name + "**" + str(exponent)
 
 
 def combine_m_match(matchobj):

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/2022-11-14-QED-DataPreparation/source/ExpressionsTokensCombiner.py

@@ -13,14 +13,6 @@ def fix_i_match(matchobj):
     """
     match = matchobj.group(0)
     return match.replace("i", "I")
-    # if int(match[1]) % 2 != 0:
-    #     print("asdf")
-    # exponent = int(match[1]) // 2
-    # m, m_name = match[0].split("_")
-    # if exponent == 1:
-    #     return m+"2"+m_name
-    # else:
-    #     return m+"2"+m_name + "**" + str(exponent)
 
 
 def combine_m_match(matchobj):

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/ampl_tree/source/ampl_to_tree.py

@@ -591,14 +591,6 @@ def is_index(s: str):
 
 
 if __name__ == "__main__":
-    # with open(sqampls_file) as f:
-    #     ampls = f.readlines()
-    #
-    # print(ampls[0])
-    # print(ampls[100])
-    # print(ampls[-1])
-    # print(ampls[-20])
-
     with open(ampls_raw_file) as f:
         ampls_raw = f.readlines(100000)
         ampls_raw = [a[:-1] for a in ampls_raw]

SYMBA_HEP/SYMBAHEP_Transformers_Abdulhakim_Alnuqaydan+Marco_Knipfer/data_preprocessing/tree/sympy_to_tree.py

@@ -125,13 +125,6 @@ def tree_to_sympy(tree, expression=None):
     else:
         node = tree._label
         op = operators_inv[node]
-        # num_args = operators_nargs[node]
-        # if num_args != len(tree):
-        #     print("num args not len(tree):")
-        #     ic(num_args)
-        #     ic(len(tree))
-        #     ic(tree)
-        # assert num_args == len(tree)
         return op(*[tree_to_sympy(t) for t in tree])
     return 0
 

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/DPOFineTuning/Tokenizers.py

@@ -282,15 +282,12 @@ def decode_float(self,seq):
                 if x[0] not in ["-", "+", "E", "N"]:
                     return np.nan
             try:
-#                 print(val)
                 sign = 1 if val[0] == "+" else -1
                 mant = ""
                 for x in val[1:-1]:
                     mant += x[1:]
                 mant = int(mant)
-#                 print(mant)
                 exp = int(val[-1][1:])
-#                 print(exp)
                 value = sign * mant * (10 ** exp)
                 value = float(value)
             except Exception:

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/DPOFineTuning/model.py

@@ -89,11 +89,8 @@ def __init__(self,
         self.register_buffer('pos_embedding_1', self.pos_embedding)
 
     def forward(self, token_embedding):
-#         print(token_embedding.shape)
         token_embedding = token_embedding.to('cuda')
         self.pos_embedding = self.pos_embedding.to('cuda')
-#         token_embedding = token_embedding
-#         self.pos_embedding = self.pos_embedding
         return self.dropout(token_embedding + self.pos_embedding[:,:token_embedding.size(1), :])
 
 

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/PIGP/Tokenizers.py

@@ -260,15 +260,12 @@ def decode_float(self,seq):
                 if x[0] not in ["-", "+", "E", "N"]:
                     return np.nan
             try:
-#                 print(val)
                 sign = 1 if val[0] == "+" else -1
                 mant = ""
                 for x in val[1:-1]:
                     mant += x[1:]
                 mant = int(mant)
-#                 print(mant)
                 exp = int(val[-1][1:])
-#                 print(exp)
                 value = sign * mant * (10 ** exp)
                 value = float(value)
             except Exception:

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/PIGP/model.py

@@ -89,11 +89,8 @@ def __init__(self,
         self.register_buffer('pos_embedding_1', self.pos_embedding)
 
     def forward(self, token_embedding):
-#         print(token_embedding.shape)
         token_embedding = token_embedding.to('cuda')
         self.pos_embedding = self.pos_embedding.to('cuda')
-#         token_embedding = token_embedding
-#         self.pos_embedding = self.pos_embedding
         return self.dropout(token_embedding + self.pos_embedding[:,:token_embedding.size(1), :])
 
 

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/SymbolicDPO/Tokenizers.py

@@ -282,15 +282,12 @@ def decode_float(self,seq):
                 if x[0] not in ["-", "+", "E", "N"]:
                     return np.nan
             try:
-#                 print(val)
                 sign = 1 if val[0] == "+" else -1
                 mant = ""
                 for x in val[1:-1]:
                     mant += x[1:]
                 mant = int(mant)
-#                 print(mant)
                 exp = int(val[-1][1:])
-#                 print(exp)
                 value = sign * mant * (10 ** exp)
                 value = float(value)
             except Exception:

SYMBA_REG/SYMBREG_DPO+PIGP_Samyak_Jha/SymbolicDPO/model.py

@@ -88,11 +88,8 @@ def __init__(self,
         self.register_buffer('pos_embedding_1', self.pos_embedding)
 
     def forward(self, token_embedding):
-#         print(token_embedding.shape)
         token_embedding = token_embedding.to('cuda')
         self.pos_embedding = self.pos_embedding.to('cuda')
-#         token_embedding = token_embedding
-#         self.pos_embedding = self.pos_embedding
         return self.dropout(token_embedding + self.pos_embedding[:,:token_embedding.size(1), :])
 
 

SYMBA_REG/SYMBREG_GP_Aryamaan_Thakur/algorithms/gp/prepare_dataset.py

@@ -27,20 +27,18 @@ def get_args_parser():
 #    """
 #    total = len(list(iterable))
 #    # Progress Bar Printing Function
-#    print(total)
 #    def printProgressBar (iteration):
 #        percent = ("{0:." + str(decimals) + "f}").format(100 * (iteration / float(total)))
 #        filledLength = int(length * iteration // total)
 #        bar = fill * filledLength + '-' * (length - filledLength)
-#        print(f'\rProgress |{bar}| {percent}% Complete', end = printEnd)
 #    # Initial Call
 #    printProgressBar(0)
 #    # Update Progress Bar
 #    for i, item in enumerate(iterable):
 #        yield item
 #        printProgressBar(i + 1)
 #    # Print New Line on Complete
-#    print()
+
 
 def main(args):
     df = pd.read_csv(args.dataframe_path)

SYMBA_REG/SYMBREG_GP_Aryamaan_Thakur/algorithms/hybrid/custom_gp.py

@@ -153,7 +153,6 @@ def get_initial_population(pop_size, candidates, pset, toolbox):
         for i in range(pop_size - len(population)):
             random_candidate = toolbox.individual()
             population.append(random_candidate)
-        #print(population)
         return population
 
     @staticmethod

SYMBA_REG/SYMBREG_GP_Aryamaan_Thakur/algorithms/hybrid/hybrid.py

@@ -46,7 +46,6 @@ def validate_expressions(self, expressions, num_vars):
         for expression in expressions:
             try:
                 expression = expression.type(torch.long)[0].tolist()
-                # print(f"{expression=}")
                 expression = self.tokenizer.reverse_tokenize([expression[1:-1]])[0]
                 #expression = prefix_to_sympy(expressions)
                 if expression not in valid:
@@ -98,10 +97,8 @@ def get_candidate_equations_single(self, X, y):
         num_vars = X.shape[1]
         x, num_array = self.format_data_for_transformer(X, y)
         expressions = self.generate_expressions(x, num_array)
-        #print(f"{expressions[0]=}")
 
         expressions = self.validate_expressions(expressions, num_vars)
-        #print(f"{expressions[0]=}")
         candidates = []
         for expression in expressions:
             try:

SYMBA_REG/SYMBREG_GP_Aryamaan_Thakur/algorithms/xval_transformers/engine/trainer.py

@@ -276,8 +276,6 @@ def test_seq_acc(self):
             y_pred = predictor.predict(src[0].unsqueeze(0), num_array[0].unsqueeze(0)) #only one example from each batch
             y_preds.append(y_pred.cpu().numpy())
             y_true.append(np.trim_zeros(tgt[0]))
-            # print("pred", y_pred.cpu().tolist())
-            # print("true", y_true[-1].tolist())
 
         test_accuracy_seq = sequence_accuracy(y_true, y_preds)
         f= open(os.path.join(self.logs_dir, "score.txt"),"w+")

SYMBA_REG/SYMBREG_GP_Aryamaan_Thakur/algorithms/xval_transformers/model/seq2seq.py

@@ -47,7 +47,6 @@ def __init__(self, vocab_size: int, emb_size):
 
     def forward(self, tokens, num_array):
         out = self.embedding(tokens.long()) * math.sqrt(self.emb_size)
-        # print("embeds", out.shape)
         out = self.layer_norm(out)
         out *= num_array.unsqueeze(-1)
         return out
@@ -68,15 +67,12 @@ def forward(self, tokens, num_array):
         #dims = torch.tensor(out.size(1)*out.size(2)*out.size(3))
         #mag_norm = 5/torch.sqrt(dims)
         #out += torch.zeros_like(out).uniform_(-mag_norm, mag_norm)
-        #print("embed", out.shape)
-        #print("num", num_array.shape)
         bs, n = out.shape[0], out.shape[1]
         out *= num_array.unsqueeze(-1)
         out = out.view(bs, n, -1)
         out = self.activation(self.fc1(out))
         out = self.dropout(out)
         out = self.fc2(out)
-        #print("out", out.shape)
         return out
 
 class Model(nn.Module):