independent km caches for distinct cases; clear the Aesara cache if it's large

Makefile

@@ -7,14 +7,20 @@ compile:
 # The reconstruction/ecoli/dataclasses/process/*.py files were generated by
 #  write_ode_file.py in Parca code.
 # Fireworks writes launcher_20* and block_20*.
+# `aesara-cache clear` clears the current $(aesara-cache) cache, not others, and
+# leaves 3 small *.delete.me dirs for later in case they're in use, esp. on NFS.
 clean:
-	rm -fr fixtures
+	rm -fr fixtures cache
 	(cd reconstruction/ecoli/dataclasses/process && rm -f equilibrium_odes.py two_component_system_odes*.py)
 	find . -name "*.pyc" -exec rm -rf {} \;
 	find . -name "*.o" -exec rm -fr {} \;
 	find . -name "*.so" -exec rm -fr {} \;
 	rm -fr build
 	rm -fr launcher_20* block_20*
+	if [ "`aesara-cache | xargs du -sm | cut -f1`" -gt 30 ]; then \
+		echo "Clearing the aesara-cache since it's larger than threshold."; \
+		aesara-cache clear; \
+	fi
 
 # Delete just the *.so libraries then (re)compile them.
 # This is useful when switching to a different Python virtualenv.

reconstruction/ecoli/fit_sim_data_1.py

@@ -5,8 +5,7 @@
 TODO: functionalize so that values are not both set and returned from some methods
 """
 
-from __future__ import absolute_import, division, print_function
-
+import binascii
 import functools
 import itertools
 import os
@@ -26,7 +25,6 @@
 from wholecell.containers.bulk_objects_container import BulkObjectsContainer
 from wholecell.utils import filepath, parallelization, units
 from wholecell.utils.fitting import normalize, masses_and_counts_for_homeostatic_target
-from wholecell.utils import parallelization
 
 
 # Fitting parameters
@@ -3164,6 +3162,13 @@ def calculateRnapRecruitment(sim_data, cell_specs):
 		}
 
 
+def crc32(arr: np.ndarray) -> int:
+	"""Return a CRC32 checksum of an ndarray."""
+	shape = str(arr.shape).encode()
+	values = arr.tobytes()
+	return binascii.crc32(shape + values)
+
+
 def setKmCooperativeEndoRNonLinearRNAdecay(sim_data, bulkContainer):
 	"""
 	Fits the affinities (Michaelis-Menten constants) for RNAs binding to endoRNAses.
@@ -3312,13 +3317,12 @@ def setKmCooperativeEndoRNonLinearRNAdecay(sim_data, bulkContainer):
 				alpha
 			)
 
+	# The checksum in the filename picks independent caches for distinct cases
+	# such as different Parca options or Parca code in different git branches.
+	# `make clean` will delete the cache files.
 	needToUpdate = False
-	fixturesDir = filepath.makedirs(filepath.ROOT_PATH, "fixtures", "endo_km")
-	# Numpy 'U' fields make these files incompatible with older code, so change
-	# the filename. No need to make files compatible between Python 2 & 3; we'd
-	# have to set the same protocol version and set Python 3-only args like
-	# encoding='latin1'.
-	km_filepath = os.path.join(fixturesDir, 'km{}.cPickle'.format(sys.version_info[0]))
+	cache_dir = filepath.makedirs(filepath.ROOT_PATH, "cache")
+	km_filepath = os.path.join(cache_dir, f'parca-km-{crc32(Kmcounts)}.cPickle')
 
 	if os.path.exists(km_filepath):
 		with open(km_filepath, "rb") as f: