def generate_binwgt_run_files(xfn, yfn, alpha, _w, fade, cats, cat_lookup,
all_qids, users, stretches, passrates, passquals,
levels, mcmcdiffs, cat_ixs, n_classes):
X_file = open(xfn, "w")
y_file = open(yfn, "w")
n_features = len(cats)
print("using n_features=", n_features)
run_ct = 0
X = numpy.zeros(
shape=n_features) #init'se a new feature vector w same width as all_X
print("Generating files for {} users...".format(len(users)))
for u in users:
#print("user = "******"../by_user/{}.txt".format(u), header=None)
runs = extract_runs_w_timestamp(attempts)
for run in runs:
run_ct += 1
ts, q, n_atts, n_pass = run
qt = q.replace("|", "~")
lev = levels[qt]
if lev == 0:
continue
qdiff = calc_qdiff(qt,
passrates,
stretches,
levels,
mcmcdiffs,
mode=_w)
catix = cat_ixs[cat_lookup[qt]]
qenc = numpy.zeros(shape=qenc_width)
qenc[catix] = 1.0
X_file.write(",".join([str(x) for x in X]) + "," +
",".join([str(e) for e in qenc]) + "\n")
X = X * fade
if (n_pass > 0):
if n_classes == 2:
y = 0
else:
y = (-1 if n_atts == 1 else 0)
X[catix] = (1.0 - alpha) * X[catix] + alpha * 1.0
else:
y = 1
y_file.write(str(y) + "\n")
X_file.flush()
y_file.flush()
X_file.close()
y_file.close()
print(n_users, "users", run_ct, "runs", run_ct / float(n_users), "rpu")
#exp_cols = [x+"_P" for x in all_qids]
#exp_cols = [x+"_X" for x in all_qids] # + [x+"_N" for x in all_qids] + [x+"_P" for x in all_qids] + ["total_qs"]
exp_mx = pandas.DataFrame( index=users, columns=exp_cols )#, columns = questions)
print("Created exp_mx of shape:", exp_mx.shape)
#cnt_mx = pandas.DataFrame( index=users, columns=list(levels) )#, columns = questions)
#exp_mx[:] = -1.0
#exp_mx = numpy.zeros((len(users),len(questions)))
exp_mx.fillna(0.0, inplace=True)
#cnt_mx.fillna(0.0, inplace=True)
for i,u in enumerate(users):
# print(u,"...")
uqatts = pandas.read_csv(base+"by_user/{}.txt".format(u), header=None)
runs = extract_runs_w_timestamp(uqatts)
for run in runs:
ts,q,n_atts,n_pass= run
q = q.replace("|","~")
# L = lev_lookup[q]
c = cat_lookup[q]
exp_mx.loc[u,c+"_X"] += 1.0
# exp_mx.loc[u,c+"_N"] += n_atts
# exp_mx.loc[u,c+"_P"] += n_pass#/n_atts
# exp_mx.loc[u,"total_qs"] += 1.0
# if(p>0):
# exp_mx.loc[u,c] += 1.0
print(u,"done")
exp_mx.fillna(0.0, inplace=True)
#print(exp_mx.shape)
def generate_run_files(alpha, _featureset_to_use, _w, fade, cats, cat_lookup, all_qids, users, stretches, passrates, passquals, levels, mcmcdiffs, cat_ixs):
stem = _featureset_to_use+"_"+str(alpha) + "_" + str(fade) + "_" + _w
x_filename= stem+"_X.csv"
y_filename= stem+"_y.csv"
X_file = open(stem+"_X.csv","w")
y_file = open(stem+"_y.csv","w")
n_features = len(cats)
# all_X = numpy.zeros(shape=(0,n_features))
print("using n_features=", n_features)
# tmx = numpy.loadtxt("../mcmc/X.csv", delimiter=",") # load the prob transition mx
# qf = open("../mcmc/obsqs.txt")
# qindex = [rec.split(",")[0] for rec in qf.read().splitlines()]
# qf.close()
#
# print(tmx.shape[0], len(qindex))
# assert tmx.shape[0] == len(qindex)
# print("loaded transition data")
run_ct= 0
X = numpy.zeros(shape=n_features) #init'se a new feature vector w same width as all_X
print("Generating files for {} users...".format(len(users)))
for u in users:
print("user = "******"../by_user/{}.txt".format(u), header=None)
runs = extract_runs_w_timestamp(attempts)
for run in runs:
run_ct+=1
ts, q, n_atts, n_pass = run
qt = q.replace("|","~")
lev = levels[qt]
if lev<1:
continue
# qdiff = calc_qdiff(qt, passrates, stretches, levels, mcmcdiffs, mode=_w)
catix = cat_ixs[ cat_lookup[qt] ]
passrate = passrates[qt]
qpassqual = passquals[qt]
stretch = stretches[qt]
mcmc = mcmcdiffs[qt] if qt in mcmcdiffs else 0
# mcmc = 0
# if(n_pass > 0):
# tailix = qindex.index(qt)
# headix = qindex.index(qt)
# mcmc = tmx[headix, tailix]
# print ("mcmc = ",mcmc)
#print(qindex)
qenc = numpy.zeros(shape=qenc_width)
# qenc[:] = 0.0 #reset question encoding
q_weight = 1.0
if _w == DW_NATTS or _w == DW_STRETCH:
q_weight = stretch
elif _w == DW_PASSRATE:
q_weight = passrate
elif _w == DW_LEVEL:
q_weight = 1+lev
elif _w == DW_MCMC:
q_weight = mcmc
qenc[catix] = q_weight # set the next q category and diff
X_file.write(",".join([str(x) for x in X])+","+",".join([str(e) for e in qenc])+"\n")
X = X * fade
a_weight = 1.0
if _w == DW_BINARY:
a_weight = 1.0
elif _w == DW_NATTS:
a_weight = n_atts
elif _w == DW_NO_WEIGHT:
a_weight = 1.0 / n_atts
elif _w == DW_PASSRATE:
a_weight = passrate / n_atts
elif _w == DW_STRETCH:
a_weight = stretch / n_atts
elif _w == DW_MCMC:
a_weight = mcmc / n_atts
elif _w == DW_LEVEL:
a_weight = (1+lev) / n_atts
if (n_pass>0):
if n_classes==2:
y = 0
else:
y = (-1 if n_atts==1 else 0)
X[catix] = (1.0-alpha)*X[catix] + alpha*a_weight
else:
y = 1
#X[catix] = 0
#X[catix] = retain*X[catix] -(1-retain)*upd
y_file.write(str(y)+"\n")
X_file.flush()
y_file.flush()
X_file.close()
y_file.close()
print(n_users, "users", run_ct,"runs", run_ct/float(n_users), "rpu")
return x_filename,y_filename
def profile_student_irt(u, ass_ts, cats, cat_lookup, cat_ixs, levels, concepts_all):
#load student's files
base = "../../../isaac_data_files/"
#cats, cat_lookup, all_qids, users, _stretches_, levels, cat_ixs = init_objects(-1, path=base, seed=666)
df = pd.read_csv(base + "hwgen1.csv", index_col=0, header=0)
#runs = open(base + "by_runs/{}.txt".format(u)).readlines()
fname = base+"by_user/{}.txt".format(u)
try:
attempts = pd.read_csv(fname, header=None)
except FileNotFoundError:
return []
runs = extract_runs_w_timestamp(attempts)
u_run_ct = len(runs)
all_zero_level = True
run_ct=0
irts = {}
subj_irts = {}
print("num of runs", len(runs))
cat_levels = [0] * len(cats)
concept_levels = [0] * 100
for run_ix, run in enumerate(runs):
run_ct += 1
ts, q, n_atts, n_pass = run
ts = pd.to_datetime(ts)
# print("rum", run_ct, ts)
if ts > ass_ts:
break
# print(ts, "<=", ass_ts)
qt = q.replace("|","~")
cat = cat_lookup[qt]
lev = levels[qt]
if q not in df.index:
continue
concepts_raw = df.loc[q, "related_concepts"]
concepts = eval(concepts_raw) if not pd.isna(concepts_raw) else []
for c in concepts:
if c not in irts:
print("at run", run_ct, "new irt engine for", u,c)
irts[c] = IRTEngine()
else:
irt = irts[c]
irt.curr_theta = irt.update(lev, (n_pass > 0))
irts[c] = irt
print(u, c,"history =", irt.history)
print("at run", run_ct, "irt update", u, c, irt.curr_theta)
if cat not in subj_irts:
subj_irts[cat] = IRTEngine()
else:
subj_irts[cat].curr_theta = subj_irts[cat].update(lev, (n_pass > 0))
print("at run", run_ct, "irt cat update", u, cat, subj_irts[cat].curr_theta)
for s in subj_irts:
catix = cat_ixs[cat]
theta = subj_irts[s].curr_theta
print(s,"=",theta)
cat_levels[catix]=theta
if irts:
print("\nConcept level abilities:")
for c in irts:
theta = irts[c].curr_theta
print(c,"=",theta)
conix = concepts_all.index(c)
concept_levels[conix]=theta
# concatd = cat_levels + concept_levels
concatd = concept_levels
print("Profile for user {}: {}".format(u, concatd))
if concatd==[]:
print("empty")
return concatd
cqmx = np.zeros(shape=n_qids)
csqmx = np.zeros(shape=n_qids)
cfqmx = np.zeros(shape=n_qids)
usersf = open("direct_mcmc_users.txt", "w")
eps_cnt = 0
qenc = np.zeros(shape=len(cats))
for u in users:
qenc[:, :] = 0.0
xp = 0
eps_cnt += 1
print("user = "******"\n")
attempts = pd.read_csv("../by_user/{}.txt".format(u), header=None)
runs = extract_runs_w_timestamp(attempts)
for run in runs[0:10]:
ts, q, n_atts, n_pass = run
q = q.replace("|", "~")
qix = all_qids.index(q)
xp += 1
qmx[qix] += qenc
cqmx[qix] += 1
if n_pass > 0:
sqmx[qix] += qenc
csqmx[qix] += 1
else:
fqmx[qix] += qenc
cfqmx[qix] += 1
catix = cat_ixs[cat_lookup[q]]
qenc[catix] += 1
end = dateutil.parser.parse(uqatts.iloc[-1, 0])
if start < target <= end:
active_users.append(u)
exp_mx_lookup[target] = numpy.zeros(shape=(len(active_users),
len(all_qids)))
print("Created exp_mx of shape:", exp_mx_lookup[target].shape)
for uix, u in enumerate(active_users):
X = numpy.zeros(shape=(1,
len(all_qids))) #, dtype=numpy.int32)
uqatts = pandas.read_csv(filebase + "by_user/{}.txt".format(u),
header=None)
uqatts[0] = pandas.to_datetime(uqatts[0])
uqatts = uqatts[uqatts[0] <= target]
# print("size for {}".format(target), uqatts.size)
for run in extract_runs_w_timestamp(uqatts):
ts, q, n_atts, n_pass = run
# q = run[0]
# n_atts = run[1]
# n_pass = run[2]
q = q.replace("|", "~")
# L = lev_lookup[q]
c = cat_lookup[q]
#decay the user's previous career
exp_mx = exp_mx_lookup[target]
#exp_mx[uix] = decay * exp_mx[uix]
if (n_atts > 0):
qix = q_ixs[q]
exp_mx[uix, qix] = 1.0
# else:
# qix = q_ixs[q]