def find_jobs_torun(self, max_njobs):
"""
Find entries whose results have not been yet calculated.
Args:
select_formulas:
"""
jobs, got = [], 0
for struct_type, formula, data in self.iter_struct_formula_data():
if got == max_njobs: break
if data in ("scheduled", "failed"): continue
if data is None:
symbols = list(set(species_from_formula(formula)))
pseudos = self.table.pseudos_with_symbols(symbols)
job = dict2namedtuple(formula=formula,
struct_type=struct_type,
pseudos=pseudos)
self[struct_type][formula] = "scheduled"
jobs.append(job)
got += 1
# Update the database.
if jobs: self.json_write()
return jobs
def plot_errors_for_elements(self, ax=None, **kwargs):
"""
Plot the relative errors associated to the chemical elements.
"""
dict_list = []
for idx, row in self.iterrows():
rerr = 100 * (row["this"] - row["ae"]) / row["ae"]
for symbol in set(species_from_formula(row.formula)):
dict_list.append(dict(
element=symbol,
rerr=rerr,
formula=row.formula,
struct_type=row.struct_type,
))
frame = DataFrame(dict_list)
order = sort_symbols_by_Z(set(frame["element"]))
#print_frame(frame)
import seaborn as sns
ax, fig, plt = get_ax_fig_plt(ax=ax)
# Draw violinplot
#sns.violinplot(x="element", y="rerr", order=order, data=frame, ax=ax, orient="v")
# Box plot
ax = sns.boxplot(x="element", y="rerr", data=frame, ax=ax, order=order, whis=np.inf, color="c")
# Add in points to show each observation
sns.stripplot(x="element", y="rerr", data=frame, ax=ax, order=order,
jitter=True, size=5, color=".3", linewidth=0)
sns.despine(left=True)
ax.set_ylabel("Relative error %")
ax.grid(True)
return fig
def select_bad_guys(self, reltol=0.4):
new = self[abs(100 * (self["this"] - self["ae"]) / self["ae"]) > reltol].copy()
new["rel_err"] = 100 * (self["this"] - self["ae"]) / self["ae"]
new.__class__ = self.__class__
count = Counter()
for idx, row in new.iterrows():
for symbol in set(species_from_formula(row.formula)):
count[symbol] += 1
return new, count
def plot_errors_for_elements(self, ax=None, **kwargs):
"""
Plot the relative errors associated to the chemical elements.
"""
dict_list = []
for idx, row in self.iterrows():
rerr = 100 * (row["this"] - row["ae"]) / row["ae"]
for symbol in set(species_from_formula(row.formula)):
dict_list.append(dict(
element=symbol,
rerr=rerr,
formula=row.formula,
struct_type=row.struct_type,
))
frame = DataFrame(dict_list)
order = sort_symbols_by_Z(set(frame["element"]))
#print_frame(frame)
import seaborn as sns
ax, fig, plt = get_ax_fig_plt(ax=ax)
# Draw violinplot
#sns.violinplot(x="element", y="rerr", order=order, data=frame, ax=ax, orient="v")
# Box plot
ax = sns.boxplot(x="element", y="rerr", data=frame, ax=ax, order=order, whis=np.inf, color="c")
# Add in points to show each observation
sns.stripplot(x="element", y="rerr", data=frame, ax=ax, order=order, hue='struct_type',
# jitter=True, size=5, color=".3", linewidth=0)
jitter=0, size=4, color=".3", linewidth=0, palette=sns.color_palette("muted"))
sns.despine(left=True)
ax.set_ylabel("Relative error %")
labels = ax.get_xticklabels()
ticks = ax.get_xticks()
ticks1 = range(min(ticks), max(ticks)+1, 2)
ticks2 = range(min(ticks) + 1, max(ticks)+1, 2)
labels1 = [labels[i].get_text() for i in ticks1]
labels2 = [labels[i].get_text() for i in ticks2]
# ax.tick_params(which='both', direction='out')
#ax.set_ylim(-1, 1)
ax.set_xticks(ticks1)
ax.set_xticklabels(labels1, rotation=90)
ax2 = ax.twiny()
ax2.set_zorder(-1)
ax2.set_xticks(ticks2)
ax2.set_xticklabels(labels2, rotation=90)
ax2.set_xlim(ax.get_xlim())
ax.grid(True)
return fig
def select_symbol(self, symbol):
"""
Extract the rows whose formula contain the given element symbol.
Return new `GbrvCompoundDataFrame`.
"""
rows = []
for idx, row in self.iterrows():
if symbol not in species_from_formula(row.formula): continue
rows.append(row)
return self.__class__(rows)
def select_symbol(self, symbol):
"""
Extract the rows whose formula contain the given element symbol.
Return new `GbrvCompoundDataFrame`.
"""
rows = []
for idx, row in self.iterrows():
if symbol not in species_from_formula(row.formula): continue
rows.append(row)
return self.__class__(rows)
def select_bad_guys(self, reltol=0.4, struct_type=None):
"""Return new frame with the entries whose relative errors is > reltol."""
new = self[abs(100 * (self["this"] - self["ae"]) / self["ae"]) > reltol].copy()
new["rel_err"] = 100 * (self["this"] - self["ae"]) / self["ae"]
if struct_type is not None:
new = new[new.struct_type == struct_type]
new.__class__ = self.__class__
count = Counter()
for idx, row in new.iterrows():
for symbol in set(species_from_formula(row.formula)):
count[symbol] += 1
new.symbol_counter = count
return new
def select_bad_guys(self, reltol=0.4, struct_type=None):
"""Return new frame with the entries whose relative errors is > reltol."""
new = self[abs(100 * (self["this"] - self["ae"]) /
self["ae"]) > reltol].copy()
new["rel_err"] = 100 * (self["this"] - self["ae"]) / self["ae"]
if struct_type is not None:
new = new[new.struct_type == struct_type]
new.__class__ = self.__class__
count = Counter()
for idx, row in new.iterrows():
for symbol in set(species_from_formula(row.formula)):
count[symbol] += 1
new.symbol_counter = count
return new
def find_jobs_torun(self, max_njobs):
"""
Find entries whose results have not been yet calculated.
Args:
select_formulas:
"""
jobs, got = [], 0
for struct_type, formula, data in self.iter_struct_formula_data():
if got == max_njobs: break
if data in ("scheduled", "failed"): continue
if data is None:
symbols = list(set(species_from_formula(formula)))
pseudos = self.table.pseudos_with_symbols(symbols)
job = dict2namedtuple(formula=formula, struct_type=struct_type, pseudos=pseudos)
self[struct_type][formula] = "scheduled"
jobs.append(job)
got += 1
# Update the database.
if jobs: self.json_write()
return jobs
def plot_errors_for_elements(self, ax=None, **kwargs):
"""
Plot the relative errors associated to the chemical elements.
"""
dict_list = []
for idx, row in self.iterrows():
rerr = 100 * (row["this"] - row["ae"]) / row["ae"]
for symbol in set(species_from_formula(row.formula)):
dict_list.append(
dict(
element=symbol,
rerr=rerr,
formula=row.formula,
struct_type=row.struct_type,
))
frame = DataFrame(dict_list)
order = sort_symbols_by_Z(set(frame["element"]))
#print_frame(frame)
import seaborn as sns
ax, fig, plt = get_ax_fig_plt(ax=ax)
# Draw violinplot
#sns.violinplot(x="element", y="rerr", order=order, data=frame, ax=ax, orient="v")
# Box plot
ax = sns.boxplot(x="element",
y="rerr",
data=frame,
ax=ax,
order=order,
whis=np.inf,
color="c")
# Add in points to show each observation
sns.stripplot(
x="element",
y="rerr",
data=frame,
ax=ax,
order=order,
hue='struct_type',
# jitter=True, size=5, color=".3", linewidth=0)
jitter=0,
size=4,
color=".3",
linewidth=0,
palette=sns.color_palette("muted"))
sns.despine(left=True)
ax.set_ylabel("Relative error %")
labels = ax.get_xticklabels()
ticks = ax.get_xticks()
ticks1 = range(min(ticks), max(ticks) + 1, 2)
ticks2 = range(min(ticks) + 1, max(ticks) + 1, 2)
labels1 = [labels[i].get_text() for i in ticks1]
labels2 = [labels[i].get_text() for i in ticks2]
# ax.tick_params(which='both', direction='out')
#ax.set_ylim(-1, 1)
ax.set_xticks(ticks1)
ax.set_xticklabels(labels1, rotation=90)
ax2 = ax.twiny()
ax2.set_zorder(-1)
ax2.set_xticks(ticks2)
ax2.set_xticklabels(labels2, rotation=90)
ax2.set_xlim(ax.get_xlim())
ax.grid(True)
return fig
def gbrv_runform(options):
"""
Run GBRV compound tests given a chemical formula.
"""
# Extract chemical symbols from formula
formula = options.formula
symbols = set(species_from_formula(formula))
# Init pseudo table and construct all possible combinations for the given formula.
table = DojoTable.from_dir(top=options.pseudo_dir,
exts=("psp8", "xml"),
exclude_dirs="_*")
pseudo_list = table.all_combinations_for_elements(symbols)
#print("Removing relativistic pseudos from list")
#pseudo_list = [plist for plist in pseudo_list if not any("_r" in p.basename for p in plist)]
# This is hard-coded since we GBRV results are PBE-only.
# There's a check between xc and pseudo.xc below.
xc = "PBE"
gbrv_factory = GbrvCompoundsFactory(xc=xc)
db = gbrv_factory.db
# Consistency check
entry = db.match_symbols(symbols)
if entry is None:
cprint("Cannot find entry for %s! Returning" % str(symbols), "red")
return 1
workdir = "GBRVCOMP_" + formula
print("Working in:", workdir)
flow = GbrvCompoundsFlow(workdir=workdir)
accuracy = "high"
for pseudos in pseudo_list:
if any(xc != p.xc for p in pseudos):
raise ValueError("Pseudos with different XC functional")
ecut = max(p.hint_for_accuracy(accuracy).ecut for p in pseudos)
pawecutdg = max(
p.hint_for_accuracy(accuracy).pawecutdg for p in pseudos)
if ecut <= 0.0: raise RuntimeError("Pseudos do not have hints")
#ecut = ecut_from_pseudos(pseudos, accuracy)
print("Adding work for pseudos:", pseudos)
print(" formula:", entry.symbol, ", structure:", entry.struct_type,
", ecut:", ecut)
work = gbrv_factory.relax_and_eos_work(accuracy,
pseudos,
entry.symbol,
entry.struct_type,
ecut=ecut,
pawecutdg=None)
flow.register_work(work)
if options.dry_run:
flow.build_and_pickle_dump()
return 0
# Run the flow with the scheduler (enable smart_io)
flow.use_smartio()
return flow.make_scheduler().start()