def inspectOne(times=0, printCorrect=True):
isAllCorrect = True
if times > 10000: # to prevent infinite loops
return
ind = np.random.randint(0, len(val.index))
ind = (val.index[ind][0], slice(None))
string_input = values_test.loc[(slice(None), ind[0]), :]
preds = model.predict(
getData(val.loc[ind], string_input, feat_x, strings_x))
preds = [np.argmax(x, axis=-1) for x in preds]
#predicted string
print("Predicted String Q", string_input[strings_x[0]][0], "from",
"-".join(str(x) for x in string_input.index.values[0]))
if (np.argmax(string_input[strings_y[0]][0], axis=-1) == preds[0]).all():
if printCorrect: print(colors.ok + '✅' + colors.close + "Segmentation")
else:
print(colors.fail + '❌' + colors.close + "Segmentation")
isAllCorrect = False
print(" T", utf2bw(string_input[strings_y[0]][0]))
print(" ", utf2bw(ctable.decode(preds[0][0], calc_argmax=False)))
# print('Q', utf2bw(pretty_join(rowx)))
rowy = dict()
for i, v in enumerate(feat_y):
rowy[v] = {"correct": val[v].loc[ind]}
res = np.zeros((SENTLEN, rowy[v]["correct"].shape[1]))
for ii, c in enumerate(preds[i + 1][0]):
res[ii, c] = 1
rowy[v]["pred"] = pd.DataFrame(res, columns=val[v].columns)
results = []
if (rowy[v]["correct"].values == rowy[v]["pred"].values).all():
if printCorrect: print(colors.ok + '✅' + colors.close + v)
else:
isAllCorrect = False
print(colors.fail + '❌' + colors.close + v, end=' ')
# results.append(colors.fail + '☒' + colors.close)
results.append('T ' + pretty_join(rowy[v]["correct"]))
results.append(pretty_join(rowy[v]["pred"]))
print(' '.join(results))
if isAllCorrect or times < 10:
print("")
inspectOne(times + 1, printCorrect)
def calc_accuracy(name, model,mydata,data_length, debug=True):
results = pd.DataFrame([], columns=[[x for x in model.output_names*3]+["agg"],[x for x in ["acc","pred","actu"]*len(model.output_names)]+["agg"]])
results.sort_index(axis=1, inplace=True)
for ite in range(data_length):
r = dict()
for i, v in enumerate(model.output_names):
if v not in strings_y:
continue
r[(v,"acc")] = (np.argmax(mydata[1][v][ite], axis=-1) == preds[i][ite]).all()
r[(v,"pred")] = utf2bw(ctable.decode(preds[i][ite], calc_argmax=False)).strip(" ") if debug else ""
r[(v,"actu")] = utf2bw(ctable.decode(mydata[1][v][ite], calc_argmax=True)).strip(" ") if debug else ""
for i, v in enumerate(model.output_names):
if v in strings_y:
continue
correct, pred = np.argmax(mydata[1][v][ite], axis=-1), preds[i][ite]
r[(v,"acc")] = (correct == pred).all()
r[(v,"actu")] = pretty_join2(correct,val[v].columns,v) if debug else ""
r[(v,"pred")] = pretty_join2(pred,val[v].columns,v) if debug else ""
r[("agg","agg")] = sum([r[(v,"acc")] for i, v in enumerate(model.output_names)]) == len(model.output_names)
results.loc[ite] = r
accuracies.loc[len(accuracies)] = [name]+list(np.average(results.filter(regex="acc|agg").values.astype(int),axis=0))
return results
ctable.decode(mydata[0][strings_x[0]][ite],
calc_argmax=True), "from",
"-".join(str(x) for x in values_test.index.values[ite]))
for ii, v in enumerate(model.output_names):
if v not in strings_y:
continue
if not (np.argmax(mydata[1][v][ite], axis=-1)
== preds[ii][ite]).all():
print('❌' + v)
isAllCorrect = False
print(
" T",
utf2bw(
ctable.decode(mydata[1][v][ite],
calc_argmax=True)))
print(
" ",
utf2bw(ctable.decode(preds[ii][ite],
calc_argmax=False)))
rowy = dict()
for ii, v in enumerate(model.output_names):
if v in strings_y:
continue
rowy[v] = {"correct": np.argmax(mydata[1][v][ite], axis=-1)}
rowy[v]["pred"] = preds[ii][ite]
results = []
# print(val[v].columns)
if not (rowy[v]["correct"] == rowy[v]["pred"]).all():
# In[207]:
# for ite in range(len(mydata[0][strings_x[0]])):
for ite in range(10):
print("\nPredicted String Q", ctable.decode(mydata[0][strings_x[0]][ite], calc_argmax=True), "from",
"-".join(str(x) for x in values_test.index.values[ite]))
for i, v in enumerate(model.output_names):
if v not in strings_y:
continue
if not (np.argmax(mydata[1][v][ite], axis=-1) == preds[i][ite]).all():
print(colors.fail + '❌' + colors.close + v)
isAllCorrect = False
print(" T", utf2bw(ctable.decode(mydata[1][v][ite], calc_argmax=True)))
print(" ", utf2bw(ctable.decode(preds[i][ite], calc_argmax=False)))
rowy = dict()
for i, v in enumerate(model.output_names):
if v in strings_y:
continue
rowy[v] = {"correct": np.argmax(mydata[1][v][ite], axis=-1)}
rowy[v]["pred"] = preds[i][ite]
results = []
# print(val[v].columns)
if not (rowy[v]["correct"] == rowy[v]["pred"]).all():
isAllCorrect = False
print(colors.fail + '❌' + colors.close + v, end=' ')
# results.append(colors.fail + '☒' + colors.close)
results.append('T ' + pretty_join2(rowy[v]["correct"],val[v].columns,v))