@Override
protected void checkMemoryFootPrint() {
if (_model._output._ntrees == 0) return;
int trees_so_far = _model._output._ntrees; // existing trees
long model_mem_size =
new ComputeModelSize(trees_so_far, _model._output._treeKeys).doAllNodes()._model_mem_size;
_model._output._treeStats._byte_size = model_mem_size;
double avg_tree_mem_size = (double) model_mem_size / trees_so_far;
Log.debug(
"Average tree size (for all classes): " + PrettyPrint.bytes((long) avg_tree_mem_size));
// all the compressed trees are stored on the driver node
long max_mem = H2O.SELF.get_max_mem();
if (_parms._ntrees * avg_tree_mem_size > max_mem) {
String msg =
"The tree model will not fit in the driver node's memory ("
+ PrettyPrint.bytes((long) avg_tree_mem_size)
+ " per tree x "
+ _parms._ntrees
+ " > "
+ PrettyPrint.bytes(max_mem)
+ ") - try decreasing ntrees and/or max_depth or increasing min_rows!";
error("_ntrees", msg);
cancel(msg);
}
}
protected void checkMemoryFootPrint() {
long mem_usage = 8 /*doubles*/ * _parms._k * _train.numCols() * (_parms._standardize ? 2 : 1);
long max_mem = H2O.SELF._heartbeat.get_free_mem();
if (mem_usage > max_mem) {
String msg =
"Centroids won't fit in the driver node's memory ("
+ PrettyPrint.bytes(mem_usage)
+ " > "
+ PrettyPrint.bytes(max_mem)
+ ") - try reducing the number of columns and/or the number of categorical factors.";
error("_train", msg);
cancel(msg);
}
}
@Override
protected void checkMemoryFootPrint() {
HeartBeat hb = H2O.SELF._heartbeat;
double p = _train.degreesOfFreedom();
long mem_usage =
(long)
(hb._cpus_allowed
* p
* p
* 8 /*doubles*/
* Math.log((double) _train.lastVec().nChunks())
/ Math.log(2.)); // one gram per core
long max_mem = hb.get_max_mem();
if (mem_usage > max_mem) {
String msg =
"Gram matrices (one per thread) won't fit in the driver node's memory ("
+ PrettyPrint.bytes(mem_usage)
+ " > "
+ PrettyPrint.bytes(max_mem)
+ ") - try reducing the number of columns and/or the number of categorical factors.";
error("_train", msg);
cancel(msg);
}
}
public Job<Frame> execImpl() {
if (integer_fraction + binary_fraction + categorical_fraction > 1)
throw new IllegalArgumentException(
"Integer, binary and categorical fractions must add up to <= 1.");
if (Math.abs(missing_fraction) > 1)
throw new IllegalArgumentException("Missing fraction must be between 0 and 1.");
if (Math.abs(integer_fraction) > 1)
throw new IllegalArgumentException("Integer fraction must be between 0 and 1.");
if (Math.abs(binary_fraction) > 1)
throw new IllegalArgumentException("Binary fraction must be between 0 and 1.");
if (Math.abs(binary_ones_fraction) > 1)
throw new IllegalArgumentException("Binary ones fraction must be between 0 and 1.");
if (Math.abs(categorical_fraction) > 1)
throw new IllegalArgumentException("Categorical fraction must be between 0 and 1.");
if (categorical_fraction > 0 && factors <= 1)
throw new IllegalArgumentException("Factors must be larger than 2 for categorical data.");
if (response_factors < 1)
throw new IllegalArgumentException(
"Response factors must be either 1 (real-valued response), or >=2 (factor levels).");
if (response_factors > 1024)
throw new IllegalArgumentException("Response factors must be <= 1024.");
if (factors > 1000000)
throw new IllegalArgumentException("Number of factors must be <= 1,000,000).");
if (cols <= 0 || rows <= 0)
throw new IllegalArgumentException("Must have number of rows > 0 and columns > 1.");
// estimate byte size of the frame
double byte_estimate =
randomize
? rows
* cols
* (binary_fraction * 1. / 8 // bits
+ categorical_fraction * (factors < 128 ? 1 : factors < 32768 ? 2 : 4)
+ integer_fraction
* (integer_range < 128
? 1
: integer_range < 32768 ? 2 : integer_range < (1 << 31) ? 4 : 8)
+ (1 - integer_fraction - binary_fraction - categorical_fraction)
* 8) // reals
+ rows * 1 // response is
: 0; // all constants - should be small
if (byte_estimate > H2O.CLOUD._memary[0].get_max_mem() * H2O.CLOUD.size())
throw new IllegalArgumentException(
"Frame is expected to require "
+ PrettyPrint.bytes((long) byte_estimate)
+ ", won't fit into H2O's memory.");
if (!randomize) {
if (integer_fraction != 0 || categorical_fraction != 0)
throw new IllegalArgumentException(
"Cannot have integer or categorical fractions > 0 unless randomize=true.");
} else {
if (value != 0)
throw new IllegalArgumentException(
"Cannot set data to a constant value if randomize=true.");
}
if (_dest == null) throw new IllegalArgumentException("Destination key cannot be null.");
FrameCreator fc = new FrameCreator(this, this._key);
start(fc, fc.nChunks() * 5);
return this;
}
/**
* Create a summary table
*
* @return
*/
TwoDimTable createSummaryTable() {
Neurons[] neurons = DeepLearningTask.makeNeuronsForTesting(this);
long byte_size = new AutoBuffer().put(this).buf().length;
TwoDimTable table =
new TwoDimTable(
"Status of Neuron Layers",
(get_params()._diagnostics ? "" : "diagnostics disabled, ")
+ (!get_params()._autoencoder ? ("predicting " + _train.lastVecName() + ", ") : "")
+ (get_params()._autoencoder
? "auto-encoder"
: _classification
? (units[units.length - 1] + "-class classification")
: "regression")
+ ", "
+ get_params()._distribution
+ " distribution, "
+ get_params()._loss
+ " loss, "
+ String.format("%,d", size())
+ " weights/biases, "
+ PrettyPrint.bytes(byte_size)
+ ", "
+ String.format("%,d", get_processed_global())
+ " training samples, "
+ "mini-batch size "
+ String.format("%,d", get_params()._mini_batch_size),
new String[neurons.length],
new String[] {
"Layer",
"Units",
"Type",
"Dropout",
"L1",
"L2",
"Mean Rate",
"Rate RMS",
"Momentum",
"Mean Weight",
"Weight RMS",
"Mean Bias",
"Bias RMS"
},
new String[] {
"int", "int", "string", "double", "double", "double", "double", "double", "double",
"double", "double", "double", "double"
},
new String[] {
"%d",
"%d",
"%s",
"%2.2f %%",
"%5f",
"%5f",
"%5f",
"%5f",
"%5f",
"%5f",
"%5f",
"%5f",
"%5f"
},
"");
for (int i = 0; i < neurons.length; ++i) {
table.set(i, 0, i + 1);
table.set(i, 1, neurons[i].units);
table.set(i, 2, neurons[i].getClass().getSimpleName());
if (i == 0) {
table.set(i, 3, neurons[i].params._input_dropout_ratio * 100);
continue;
} else if (i < neurons.length - 1) {
if (neurons[i].params._hidden_dropout_ratios == null) {
table.set(i, 3, 0);
} else {
table.set(i, 3, neurons[i].params._hidden_dropout_ratios[i - 1] * 100);
}
}
table.set(i, 4, neurons[i].params._l1);
table.set(i, 5, neurons[i].params._l2);
table.set(
i,
6,
(get_params()._adaptive_rate ? mean_rate[i] : neurons[i].rate(get_processed_total())));
table.set(i, 7, (get_params()._adaptive_rate ? rms_rate[i] : 0));
table.set(i, 8, get_params()._adaptive_rate ? 0 : neurons[i].momentum(get_processed_total()));
table.set(i, 9, mean_weight[i]);
table.set(i, 10, rms_weight[i]);
table.set(i, 11, mean_bias[i]);
table.set(i, 12, rms_bias[i]);
}
summaryTable = table;
return summaryTable;
}
