def generate_bracket_csv(aug_train_set):
df = aug_train_set[aug_train_set.bracketing_pattern == brapa]
grouped = df.groupby("tube_assembly_id")
taids = []
fixed_costs = []
var_costs = []
for taid, indices in grouped.groups.iteritems():
quantities = df.quantity[indices].values
costs = inverse_log_transform_y(df.log_cost[indices].values)
fixed_cost, var_cost, r2 = get_fixed_and_var_cost(quantities, costs)
if r2 < 0.9999:
print "{} has bad r2".format(taid)
taids.append(taid)
fixed_costs.append(fixed_cost)
var_costs.append(var_cost)
fixed_costs = np.array(fixed_costs)
fc_class = -1 * np.ones(len(taids), dtype=np.int)
adj_fixed_costs = np.zeros(len(taids))
for i, fc_val in enumerate(fc_vals):
indices = np.abs(fixed_costs - fc_val) < 0.1
fc_class[indices] = i
adj_fixed_costs[indices] = fc_val
assert np.all(np.unique(fc_class) == [0, 1, 2, 3])
adj_var_costs = np.zeros(len(taids))
for i, taid in enumerate(taids):
indices = grouped.groups[taid]
quantities = df.quantity[indices].values
costs = inverse_log_transform_y(df.log_cost[indices].values)
fixed_cost = adj_fixed_costs[i]
adj_var_costs[i] = get_var_cost_only(quantities, costs, fixed_cost)
assert np.abs(adj_var_costs[i] - var_costs[i]) < 0.01
df = pd.DataFrame(
{
"tube_assembly_id": taids,
"fixed_cost_class": fc_class,
"fixed_cost": adj_fixed_costs,
"var_cost": adj_var_costs,
}
)
df.to_csv("bracket.csv", index=False, columns=["tube_assembly_id", "fixed_cost_class", "fixed_cost", "var_cost"])
def generate_bracket_csv(aug_train_set):
df = aug_train_set[aug_train_set.bracketing_pattern == brapa]
grouped = df.groupby('tube_assembly_id')
taids = []
fixed_costs = []
var_costs = []
for taid, indices in grouped.groups.iteritems():
quantities = df.quantity[indices].values
costs = inverse_log_transform_y(df.log_cost[indices].values)
fixed_cost, var_cost, r2 = get_fixed_and_var_cost(quantities, costs)
if r2 < 0.9999:
print "{} has bad r2".format(taid)
taids.append(taid)
fixed_costs.append(fixed_cost)
var_costs.append(var_cost)
fixed_costs = np.array(fixed_costs)
fc_class = -1 * np.ones(len(taids), dtype=np.int)
adj_fixed_costs = np.zeros(len(taids))
for i, fc_val in enumerate(fc_vals):
indices = np.abs(fixed_costs - fc_val) < 0.1
fc_class[indices] = i
adj_fixed_costs[indices] = fc_val
assert np.all(np.unique(fc_class) == [0, 1, 2, 3])
adj_var_costs = np.zeros(len(taids))
for i, taid in enumerate(taids):
indices = grouped.groups[taid]
quantities = df.quantity[indices].values
costs = inverse_log_transform_y(df.log_cost[indices].values)
fixed_cost = adj_fixed_costs[i]
adj_var_costs[i] = get_var_cost_only(quantities, costs, fixed_cost)
assert np.abs(adj_var_costs[i] - var_costs[i]) < 0.01
df = pd.DataFrame({
'tube_assembly_id': taids,
'fixed_cost_class': fc_class,
'fixed_cost': adj_fixed_costs,
'var_cost': adj_var_costs,
})
df.to_csv('bracket.csv', index=False, columns=[
'tube_assembly_id', 'fixed_cost_class', 'fixed_cost', 'var_cost'])
X_train = aug_train_set
y_train = X_train.pop('log_cost')
X_test = aug_test_set
print "Predicting..."
timer = time()
y_train_pred = get_predictions('all', expert_names, base_get_indices,
aug_train_set)
train_rmsle = np.sqrt(mean_squared_error(y_train.values, y_train_pred))
print "train RMSLE", train_rmsle
y_test_pred = get_predictions('all', expert_names, base_get_indices,
aug_test_set)
timer = time() - timer
print " {} seconds elapsed".format(timer)
print "Writing output..."
timer = time()
df = pd.DataFrame()
df['cost'] = inverse_log_transform_y(y_train_pred)
df['id'] = df.index + 1
df.to_csv("train_pred.csv", index=False, columns=['id', 'cost'])
df = pd.DataFrame()
df['cost'] = inverse_log_transform_y(y_test_pred)
df['id'] = df.index + 1
df.to_csv("test_pred.csv", index=False, columns=['id', 'cost'])
timer = time() - timer
print " {} seconds elapsed".format(timer)
print "Done!"
X_train = aug_train_set
y_train = X_train.pop('log_cost')
X_test = aug_test_set
print "Predicting..."
timer = time()
y_train_pred = get_predictions(
'all', ['base'], base_get_indices, aug_train_set)
train_rmsle = np.sqrt(mean_squared_error(y_train.values, y_train_pred))
print "train RMSLE", train_rmsle
y_test_pred = get_predictions(
'all', ['base'], base_get_indices, aug_test_set)
timer = time() - timer
print " {} seconds elapsed".format(timer)
print "Writing output..."
timer = time()
df = pd.DataFrame()
df['cost'] = inverse_log_transform_y(y_train_pred)
df['id'] = df.index + 1
df.to_csv("train_pred.csv", index=False, columns=['id', 'cost'])
df = pd.DataFrame()
df['cost'] = inverse_log_transform_y(y_test_pred)
df['id'] = df.index + 1
df.to_csv("test_pred.csv", index=False, columns=['id', 'cost'])
timer = time() - timer
print " {} seconds elapsed".format(timer)
print "Done!"
