Develop Reference

Loading TFlite model for Inference (Python)

I'm using Tensorflow Lite to train an image classifier. I now have a bunch of *.tflite models stored, and I'm trying to write some code that allows me to pick a tflite model file, pick a dataset, and test that model on that dataset (inference).
When I train a model using:
model = image_classifier.create(trainData, validation_data=valData, shuffle=True, use_augmentation=False)
I am able to easily test this model on a test dataset right away because the model is actually stored in the variable 'model', by using:
model.evaluate_tflite('model.tflite', test_data)
or
loss, accuracy = model.evaluate(test_data)
However, if I simply want to load an already existing *.tflite model, without having trained it in the same run, I can't figure out a simple way to do that.
Following these instructions, it seems to be a lot of steps for what I'm trying to do. In other Machine Learning libraries (like PyTorch), you are able to define the model and then quickly load the saved weights and then get to testing, like:
model = models.densenet201(progress=True, pretrained=pretrained)
model.load_state_dict(torch.load("models/model.pt"))
Is there a simple way for me to initialise the model into the 'model' variable, load the saved weights from a *.tflite file, and then run inference?
Thank you for your help

A simple example of image classification:
import tensorflow as tf
import numpy as np
import cv2
class TFLiteModel:
def __init__(self, model_path: str):
self.interpreter = tf.lite.Interpreter(model_path)
self.interpreter.allocate_tensors()
self.input_details = self.interpreter.get_input_details()
self.output_details = self.interpreter.get_output_details()
def predict(self, *data_args):
assert len(data_args) == len(self.input_details)
for data, details in zip(data_args, self.input_details):
self.interpreter.set_tensor(details["index"], data)
self.interpreter.invoke()
return self.interpreter.get_tensor(self.output_details[0]["index"])
model = TFLiteModel("mobilenet_v2_1.0_224_1_default_1.tflite")
image = cv2.imread("hand_blower.png")
image = cv2.resize(image, (224, 224))
image = image.astype(np.float32)[np.newaxis]
image = (image - 127.5) / 127.5
label = model.predict(image)[0].argmax()
print(label)
Please refer to the official documentation for detailed information:
https://www.tensorflow.org/api_docs/python/tf/lite/Interpreter
The model was loaded from:
https://tfhub.dev/tensorflow/lite-model/mobilenet_v2_1.0_224/1/default/1

Get accuracy metric from leaderboard function when using H20 AI

I am running a binary classification model using H2O autoML. I have explicitly told autoML to treat this as a classification model with the following line of code.
# This line of code turns our int variable into a factor.
# This is necessary to tell H2O that we want a classification model
feature_data['Radius'] = feature_data['Radius'].asfactor()
After running H20 autoML for a minute and then using the following line of code;
lb = aml.leaderboard
lb.head()
lb.head(rows=lb.nrows) # Entire leaderboard
I got the output in the screenshot below
As you can see, the metrics used for classification are AUC and logloss but what I want to see is accuracy. What should I add to get such an output?

It doesn't look like the leaderboard allows you to sort using accuracy as a metric. The following lines of code and text have been directly taken from the documentation:
aml = H2OAutoML(max_runtime_secs = 30, sort_metric = "logloss")
For binomial classification choose between AUC, "logloss", "mean_per_class_error", "RMSE", "MSE".

you can simply assign "accuracy" as a sorting metric while building h2o aml model like:
aml = H2OAutoML(max_runtime_secs = 30, sort_metric = "accuracy")
It will publish the model based on accuracy in ascending order from top to bottom.

Problem with loading language_model_learner fastai

I have problem with fastai library. My code below:
import fastai
from fastai.text import *
import os
import pandas as pd
import fastai
from fastai import *
lab = df.columns[0]
data_lm = TextLMDataBunch.from_csv(r'/AWD', 'data.csv', label_cols = lab, text_cols = ['text'])
data_clas = TextClasDataBunch.from_csv(r'/AWD', 'data.csv', vocab = data_lm.train_ds.vocab, bs = 256,label_cols = lab, text_cols=['text'])
data_lm.save('data_lm_export.pkl')
data_clas.save('data_clas.pkl')
learn = language_model_learner(data_lm,AWD_LSTM,drop_mult = 0.3)
learn.lr_find()
learn.recorder.plot(skip_end=10)
learn.fit_one_cycle(10,1e-2,moms=(0.8,0.7))
learn.save('fit_head')
learn.load('fit_head')
My data is quite big, so each epoch in fit_one_cycle lasts about 6h. My resources enables me only to train model in SLURM JOB 70h, so my whole script will be cancelled. I wanted to divide my script into pieces and the first longest part has to learn and save fit_head. Everything was ok, and after that I wanted to load my model to train it again, but i got this error:
**RuntimeError: Error(s) in loading state_dict for SequentialRNN:
size mismatch for 0.encoder.weight: copying a param with shape torch.Size([54376, 400]) from checkpoint, the shape in current model is torch.Size([54720, 400]).
**
I have checked similar problems on github/stack posts and I tried those solutions like this below, but i cannot find anything usefull.
data_clas.vocab.stoi = data_lm.vocab.stoi
data_clas.vocab.itos = data_lm.vocab.itos
Is there any possibility to load trained model without having this issue ?

When you do learner.save() only the model weights are saved on your disk and not the model state dict which contains the model architecture information.
To train the model in a different session you must first define the model itself. Remember to use the same code to define your new model. Since your data is quite heavy as you mentioned you can use a very small subset (~16 records) of your data to create this new model and then do learn.load(model_path) and you should be able to resume training.
you can modify the training data with learn.data.train_dl = new_dl

How to use tf.maybe_batch to switch between train/val pipelines?

I have a pipeline to read train and validation datasets from tfrecords.
I build batches using tf.train.batch. During training I want to switch between training and evaluation on validation dataset.
Here is simplified snippet of code how I implement it now.
is_training_pl = tf.placeholder(tf.bool)
images_train, labels_train = tf.train.batch([img_train, label_train])
images_val, labels_val = tf.train.batch([img_val, label_val])
data = tf.cond(is_training_pl, lambda: [images_train, labels_train], lambda: [images_val, labels_val])
loss = my_model(input=data)
I know that one can do it with tf.cond, but the problem with it is that both train and val batch ops would be executed when tf.cond is called.
On github ebrevdo told (link to the comment) that it's possible to use tf.train.maybe_batch for this purpose instead, which is more efficient.
Can anyone give an example of how to use tf.train.batch in my case please?

Lasagne/Theano, problems loading pickled model

I think I'm losing my mind at this point.
I'm using Lasagne for a small convolutional neural network. It trains perfectly, I can compute the error on training and validation set as well, but I cannot save the trained model on the disk. Better, I can save it and load it, but I cannot use it to predict for new data.
This is what I do after training
model = {'network': network, 'params': get_all_params(network), 'params_values': get_all_param_values(network)}
pickle.dump(model, open('models/model_1.pkl', 'wb'), protocol=pickle.HIGHEST_PROTOCOL)
And this is what I do to load the model
with open('models/model.pkl', 'rb') as pickle_file:
model = pickle.load(pickle_file)
network = model['network']
values = model['params_values']
set_all_param_values(network, values)
T_input = T.tensor4('input', dtype='float32')
T_target = T.ivector('target')
predictions = get_output(network, deterministic=True)
loss = (cross_entropy(predictions, T_target)).mean()
acc = T.mean(T.eq(T.argmax(predictions, axis=1), T_target), dtype=config.floatX)
test_fn = function([T_input, T_target], [loss, acc])
I cannot even pass the real numpy input, that I get this error
theano.compile.function_module.UnusedInputError: theano.function was asked to create a
function computing outputs given certain inputs, but the provided input variable at index 0
is not part of the computational graph needed to compute the outputs: input.
To make this error into a warning, you can pass the parameter
on_unused_input='warn' to theano.function. To disable it completely, use
on_unused_input='ignore'.
I tried to set the parameter on_unused_input='warn' then, and this is the result
theano.gof.fg.MissingInputError: An input of the graph, used to compute (..)
was not provided and not given a value.Use the Theano flag
exception_verbosity='high',for more information on this error.

The problem is that your T_input is not tied to the input layer and hence theano can't compile it
T_input = lasagne.layers.get_all_layers(network)[0].input_var