add experimental results

EleutherAI · May 22, 2021 · e9b2edc · e9b2edc
1 parent eb7ab95
commit e9b2edc
Show file tree

Hide file tree

Showing 12 changed files with 346 additions and 557 deletions.
diff --git a/README.md b/README.md
@@ -17,10 +17,10 @@ Special emphasis is placed on reusability and ease of use within diverse pipelin
 
 length   |  sota  | **us (cpu)** |  Nx   | us (gpu) | us (hybrid) |
 ---------|--------|--------------|-------|----------|-------------|
-~114     | 2.4s   | **5.64ms**   | ~420  | 22.02ms  | 19.65ms     |
-~300     | 3.5s   | **8.29ms**   | ~422  | 26.46ms  | 23.60ms     |
-~500     | 7.5s   | **11.49ms**  | ~652  | 31.47ms  | 24.96ms     |
-~1000    | 18.66s | **16.80ms**  | ~1112 | 43.09ms  | 29.09ms     |
+~114     | 2.4s   | **5.33ms**   | ~446  | 19.93ms  | 18.41ms     |
+~300     | 3.5s   | **8.66ms**   | ~400  | 25.47ms  | 22.54ms     |
+~500     | 7.5s   | **11.22ms**  | ~651  | 30.79ms  | 24.05ms     |
+~1000    | 18.66s | **18.11ms**  | ~1030 | 44.29ms  | 30.69ms     |
 
 * **Profiler Trace (CPU)**:
 <center><img src="notebooks/experiments_manual/profiler_capture.png"></center>

diff --git a/notebooks/experiments/100_info.joblib b/notebooks/experiments/100_info.joblib
diff --git a/notebooks/experiments/200_info.joblib b/notebooks/experiments/200_info.joblib
diff --git a/notebooks/experiments/300_info.joblib b/notebooks/experiments/300_info.joblib
diff --git a/notebooks/experiments/400_info.joblib b/notebooks/experiments/400_info.joblib
diff --git a/notebooks/experiments/500_info.joblib b/notebooks/experiments/500_info.joblib
diff --git a/notebooks/experiments/600_info.joblib b/notebooks/experiments/600_info.joblib
diff --git a/notebooks/experiments/700_info.joblib b/notebooks/experiments/700_info.joblib
diff --git a/notebooks/experiments/800_info.joblib b/notebooks/experiments/800_info.joblib
diff --git a/notebooks/experiments/900_info.joblib b/notebooks/experiments/900_info.joblib
diff --git a/notebooks/experiments/logs_experiment.txt b/notebooks/experiments/logs_experiment.txt
diff --git a/notebooks/integrated_test.py b/notebooks/integrated_test.py
@@ -1,95 +1,3 @@
-<<<<<<< HEAD
-##########################
-# Clone repos with utils #
-##########################
-
-# !git clone https://github.com/hypnopump/geometric-vector-perceptron
-
-import os
-import sys
-import time
-import timeit
-import logging
-
-# science
-import numpy as np 
-import torch
-import sidechainnet
-from sidechainnet.utils.sequence import ProteinVocabulary as VOCAB
-VOCAB = VOCAB()
-
-# process
-import joblib
-
-# custom
-import mp_nerf
-
-BASE_FOLDER = "experiments/"
-
-logging.basicConfig(level=logging.DEBUG,
-                    format="%(asctime)s %(levelname)s %(threadName)s %(name)s %(message)s",
-                    # datefmt='%m-%d %H:%M',
-                    filename=BASE_FOLDER+"logs_experiment.txt",
-                    filemode="a")
-logger = logging.getLogger()
-sep = "\n\n=======\n\n"
-
-
-# begin tests
-if __name__ == "__main__":
-
-    logger.info("Loading data"+"\n")
-    lengths = [100, 200, 300, 400, 500, 600, 700, 800, 900]# [::-1]
-    try: 
-        "a"+9
-        # skip
-        dataloaders_ = sidechainnet.load(casp_version=7, with_pytorch="dataloaders", batch_size=2)
-        logger.info("Data has been loaded"+"\n"+sep)
-        stored  = [ mp_nerf.utils.get_prot(dataloader_=dataloaders_, 
-                                           vocab_=VOCAB, 
-                                           min_len=desired_len+5, 
-                                           max_len=desired_len+60) for desired_len in lengths ]
-        joblib.dump(stored, BASE_FOLDER[:-1]+"_manual/analyzed_prots.joblib")
-    except: 
-        stored = joblib.load(BASE_FOLDER[:-1]+"_manual/analyzed_prots.joblib")
-        logger.info("Data has been loaded"+"\n"+sep)
-
-    logger.info("Assessing lengths of: "+str([len(x[0]) for x in stored])+"\n")
-
-    cuda_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    for device in ["cpu", cuda_device]:
-
-        logger.info("Preparing speed tests: for device "+repr(device)+"\n")
-
-        for i,desired_len in enumerate(lengths):
-
-            seq, int_seq, true_coords, angles, padding_seq, mask, pid = stored[i]
-            scaffolds = mp_nerf.proteins.build_scaffolds_from_scn_angles(seq, angles.to(device))
-
-            logger.info("Assessing the speed of folding algorithm at length "+str(len(seq))+"\n")
-
-            logger.info( str( timeit.timeit('mp_nerf.proteins.protein_fold(**scaffolds, device=device, hybrid=True)',
-            	                             globals=globals(), number=1000) )+" for 1000 calls" )
-
-            logger.info("Saving the related information at {0}{1}_info.joblib\n".format(
-                        BASE_FOLDER, desired_len))
-            joblib.dump({"seq": seq, 
-                         "true_coords": true_coords,
-                         "angles": angles,
-                         "padding_seq": padding_seq,
-                         "mask": mask,
-                         "pid": pid, 
-                         "padding_stripped": True}, BASE_FOLDER+str(desired_len)+"_info.joblib")
-            logger.info(sep)
-
-    logger.info("Execution has finished\n")
-
-
-
-
-
-
-=======
 ##########################
 # Clone repos with utils #
 ##########################
@@ -128,42 +36,35 @@
 
 # begin tests
 if __name__ == "__main__":
-    "a"+9
+
     logger.info("Loading data"+"\n")
-    # adapt desired lengths to previous experiments results
-    # lengths = [100, 200, 300, 400, 500, 600, 700, 800, 900]# [::-1]
-    lengths = [[131, 150],
-               [200, 250],
-               [331, 351],
-               [400, 450],
-               [500, 550],
-               [600, 650],
-               [700, 780],
-               [800, 900],
-               [905, 1070]]# [::-1]
+    lengths = [100, 200, 300, 400, 500, 600, 700, 800, 900]# [::-1]
     try: 
+        "a"+9
         # skip
+        dataloaders_ = sidechainnet.load(casp_version=7, with_pytorch="dataloaders", batch_size=2)
         logger.info("Data has been loaded"+"\n"+sep)
-        stored  = [ ]
-        for i,(desired_len, upper_len) in enumerate(lengths[::-1]): 
-            dataloaders_ = sidechainnet.load(casp_version=7, with_pytorch="dataloaders", batch_size=2)
-            stored.append(mp_nerf.utils.get_prot(dataloader_=dataloaders_, 
-                                                 vocab_=VOCAB, 
-                                                 min_len=desired_len, 
-                                                 max_len=upper_len))
-        stored = stored[::-1]
-
+        stored  = [ mp_nerf.utils.get_prot(dataloader_=dataloaders_, 
+                                           vocab_=VOCAB, 
+                                           min_len=desired_len+5, 
+                                           max_len=desired_len+60) for desired_len in lengths ]
         joblib.dump(stored, BASE_FOLDER[:-1]+"_manual/analyzed_prots.joblib")
     except: 
         stored = joblib.load(BASE_FOLDER[:-1]+"_manual/analyzed_prots.joblib")
         logger.info("Data has been loaded"+"\n"+sep)
 
     logger.info("Assessing lengths of: "+str([len(x[0]) for x in stored])+"\n")
 
-    cuda_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    for device in ["cpu", cuda_device]:
+    run_opts = [(torch.device("cpu"), False)] # tuples of (device, hybrid)
+    # add possibility for different configs
+    if torch.cuda.is_available():
+        run_opts.append( (torch.device("cuda"), True))
+        run_opts.append( (torch.device("cuda"), False))
+
 
-        logger.info("Preparing speed tests: for device "+repr(device)+"\n")
+    for device,hybrid in run_opts:
+
+        logger.info("Preparing speed tests: for device "+repr(device)+" and hybrid_opt: "+str(hybrid)+"\n")
 
         for i,desired_len in enumerate(lengths):
 
@@ -172,7 +73,7 @@
 
             logger.info("Assessing the speed of folding algorithm at length "+str(len(seq))+"\n")
 
-            logger.info( str( timeit.timeit('mp_nerf.proteins.protein_fold(**scaffolds, device=device, hybrid=True)',
+            logger.info( str( timeit.timeit('mp_nerf.proteins.protein_fold(**scaffolds, device=device, hybrid=hybrid)',
             	                             globals=globals(), number=1000) )+" for 1000 calls" )
 
             logger.info("Saving the related information at {0}{1}_info.joblib\n".format(
@@ -186,11 +87,4 @@
                          "padding_stripped": True}, BASE_FOLDER+str(desired_len)+"_info.joblib")
             logger.info(sep)
 
-    logger.info("Execution has finished\n")
-
-
-
-
-
-
->>>>>>> 4cabbe55371d6a9a7edeab1db719fa0cf8312eae
+    logger.info("Execution has finished\n")