def sensitivity_analysis_for_gradient_ascent(gc, nist, cid2pmap, max_i=250, n_begin=0):
# Run the gradient ascent algorithm, where the starting point is Alberty's table from (Mathematica 2006)
# Use DETERMINISTIC gradient ascent
# run the simulation many times, each time removing a fraction of the reactions from the training set, and
# evaluating the MSE only on that fraction (i.e. it will be the test set)
grad = GradientAscent(gc)
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.load_nist_data(nist, skip_missing_reactions=True)
res_file = open('../res/leave1out_report.csv', 'w')
csv_results = csv.writer(res_file)
csv_results.writerow(["N", "dG0_obs", "dG0_est_before", "dG0_est_after", "iterations", "reaction", "pH", "I", "T", "evaluation", "no. measurements"])
N = len(grad.data)
for n in range(n_begin, N):
(sparse_reaction, pH, I, T, evaluation, dG0_obs) = grad.data[n]
n_measurements = min([nist.cid2count[cid] for cid in sparse_reaction.keys()])
reaction_str = gc.kegg().sparse_reaction_to_string(sparse_reaction, cids=True)
sys.stderr.write("%04d) %s, %.1f, %.2f, %.1f, %s, %d, " % (n, reaction_str, pH, I, T, evaluation, n_measurements))
dG0_est_before = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
sys.stderr.write("E-before = %.2f, " % abs(dG0_obs - dG0_est_before))
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.train_rowids = range(n) + range(n+1, N)
grad.update_cache()
grad.deterministic_hill_climb(max_i=max_i, verbose=False)
dG0_est_after = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
csv_results.writerow([n, dG0_obs, dG0_est_before, dG0_est_after, reaction_str, max_i, pH, I, T, evaluation, n_measurements])
res_file.flush()
sys.stderr.write("E-after = %.2f\n" % abs(dG0_obs - dG0_est_after))
def evaluate(gc, nist, cid2pmap):
# Run the gradient ascent algorithm, where the starting point is Alberty's table from (Mathematica 2006)
# Use DETERMINISTIC gradient ascent
# run the simulation many times, each time removing a fraction of the reactions from the training set, and
# evaluating the MSE only on that fraction (i.e. it will be the test set)
grad = GradientAscent(gc)
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.load_nist_data(nist, skip_missing_reactions=True)
res_file = open('../res/evaluation_report.csv', 'w')
csv_results = csv.writer(res_file)
csv_results.writerow(
["N", "dG0_obs", "dG0_est", "reaction", "pH", "I", "T", "evaluation"])
N = len(grad.data)
for n in range(n_begin, N):
(sparse_reaction, pH, I, T, evaluation, dG0_obs) = grad.data[n]
n_measurements = min(
[nist.cid2count[cid] for cid in sparse_reaction.keys()])
reaction_str = gc.kegg().sparse_reaction_to_string(sparse_reaction,
cids=True)
dG0_est = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
csv_results.writerow([
n, dG0_obs, dG0_est, reaction_str, pH, I, T, evaluation,
n_measurements
])
res_file.flush()
def sensitivity_analysis_for_gradient_ascent(gc,
nist,
cid2pmap,
max_i=250,
n_begin=0):
# Run the gradient ascent algorithm, where the starting point is Alberty's table from (Mathematica 2006)
# Use DETERMINISTIC gradient ascent
# run the simulation many times, each time removing a fraction of the reactions from the training set, and
# evaluating the MSE only on that fraction (i.e. it will be the test set)
grad = GradientAscent(gc)
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.load_nist_data(nist, skip_missing_reactions=True)
res_file = open('../res/leave1out_report.csv', 'w')
csv_results = csv.writer(res_file)
csv_results.writerow([
"N", "dG0_obs", "dG0_est_before", "dG0_est_after", "iterations",
"reaction", "pH", "I", "T", "evaluation", "no. measurements"
])
N = len(grad.data)
for n in range(n_begin, N):
(sparse_reaction, pH, I, T, evaluation, dG0_obs) = grad.data[n]
n_measurements = min(
[nist.cid2count[cid] for cid in sparse_reaction.keys()])
reaction_str = gc.kegg().sparse_reaction_to_string(sparse_reaction,
cids=True)
sys.stderr.write(
"%04d) %s, %.1f, %.2f, %.1f, %s, %d, " %
(n, reaction_str, pH, I, T, evaluation, n_measurements))
dG0_est_before = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
sys.stderr.write("E-before = %.2f, " % abs(dG0_obs - dG0_est_before))
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.train_rowids = range(n) + range(n + 1, N)
grad.update_cache()
grad.deterministic_hill_climb(max_i=max_i, verbose=False)
dG0_est_after = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
csv_results.writerow([
n, dG0_obs, dG0_est_before, dG0_est_after, reaction_str, max_i, pH,
I, T, evaluation, n_measurements
])
res_file.flush()
sys.stderr.write("E-after = %.2f\n" % abs(dG0_obs - dG0_est_after))
def evaluate(gc, nist, cid2pmap):
# Run the gradient ascent algorithm, where the starting point is Alberty's table from (Mathematica 2006)
# Use DETERMINISTIC gradient ascent
# run the simulation many times, each time removing a fraction of the reactions from the training set, and
# evaluating the MSE only on that fraction (i.e. it will be the test set)
grad = GradientAscent(gc)
grad.cid2pmap_dict = deepcopy(cid2pmap)
grad.load_nist_data(nist, skip_missing_reactions=True)
res_file = open('../res/evaluation_report.csv', 'w')
csv_results = csv.writer(res_file)
csv_results.writerow(["N", "dG0_obs", "dG0_est", "reaction", "pH", "I", "T", "evaluation"])
N = len(grad.data)
for n in range(n_begin, N):
(sparse_reaction, pH, I, T, evaluation, dG0_obs) = grad.data[n]
n_measurements = min([nist.cid2count[cid] for cid in sparse_reaction.keys()])
reaction_str = gc.kegg().sparse_reaction_to_string(sparse_reaction, cids=True)
dG0_est = grad.reaction_to_dG0(sparse_reaction, pH, I, T)
csv_results.writerow([n, dG0_obs, dG0_est, reaction_str, pH, I, T, evaluation, n_measurements])
res_file.flush()